Artin - Grothendieck Topologies

GROTHENDIECK TOPOLOGIES
Notes on a Seminar by
M. Artin
Spring, 1962
HARVARD UNIVERSITY
Department of Mathematics
October, 2006
TABLE OF CONTENTS
TABLE OF CONTENTS
TABLE OF CONTENTS i
1 Preliminaries 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Generalities on lim . . . . . . . . . . . . . . . . . . . . . . . . . . 3
−→
1.3 Presheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Čech cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Sheaves 17
2.1 Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 The spectral sequences for Čech cohomology . . . . . . . . . . . . 25
2.4 The category T op, Leray spectral sequences . . . . . . . . . . . . 26
2.5 Inductive limits for nötherian topologies . . . . . . . . . . . . . . 33
3 The étale Grothendieck topology for schemes 37

3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 Relations with a closed subscheme . . . . . . . . . . . . . . . . . . 41
3.3 Passage to the limit . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4 Hensel rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.5 Cohomological dimension . . . . . . . . . . . . . . . . . . . . . . . 64
4 Calculations for curves and surfaces 69

4.1 General facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 Case of an algebraic curve . . . . . . . . . . . . . . . . . . . . . . 71
4.3 Local calculations in dimension 2 . . . . . . . . . . . . . . . . . . 73
i
TABLE OF CONTENTS
4.4 Digression on the Picard group of a curve over k0 {t} . . . . . . . 76

4.5 Cohomology of a curve over k0 {t} . . . . . . . . . . . . . . . . . . 82
4.6 Case of an algebraic surface . . . . . . . . . . . . . . . . . . . . . 87
BIBLIOGRAPHY 91
ii
1 PRELIMINARIES
1 Preliminaries
1.1 Introduction
Example 1.1.1. Let X be a topological space and let T be the category whose
objects are open sets of X and with Hom(U, V )consisting of the inclusion map
if U ⊂ V , empty otherwise. Then a presheaf F on X with values in a category
C is a contravariant functor F : T ◦ → C. A sheaf F is a presheaf satisfying the
following axiom: For U ∈ T , {Ui } an open covering of U the sequence
Q //
Q
F(U ) / F(Ui ) F(Ui ∩ Uj )
i i,j
is exact with the canonical maps. (We assume C has products. For the definition
of exactness, see [3] n◦ 195). Note that in T (in fact in the category of topological
spaces) Ui ∩ Uj ≈ Ui ×U Uj . Therefore,
Definition 1.1.1. A Grothendieck topology T consists of a category T and a set

Cov T of families {Ui −−φ−i→ U }i∈I of maps in T called coverings (where in each
covering the range U of the maps φi is fixed) satisfying
1. If φi is an isomorphism then {φ} ∈ Cov T ;
2. If {Ui → U } ∈ Cov T and {Vij → Ui } ∈ Cov T for each i then the family
{Vij → U } obtained by composition is in Cov T .
3. If {Ui → U } ∈ Cov T and V → U ∈ T is arbitrary then Ui ×U V exists and

{Ui ×U V → V } ∈ Cov T
We will abuse language and call T a topology.
Definition 1.1.2. Let T be a topology and C a category with products. A

presheaf on T with values in C is a functor F : T ◦ → C. A sheaf F is a presheaf
satisfying
1
1.1 INTRODUCTION
(S) If {Ui → U } ∈ Cov T then the diagram

Q //
Q
/
F(U ) i F(Ui ) i,j F(Ui ×U Uj )
is exact.
We shall restrict our attention to presheaves and sheaves with value in Ab

or Sets, (mostly Ab). Note that if F has values in Ab then F is a sheaf iff it “is”
a sheaf of sets.
Example 1.1.2. Let C be any category with fiber products. Then there is a
canonical topology TC associated to C, namely set TC = C, Cov TC = set of
families of maps {Ui → U } in C which are universal effective epimorphisms, i.e.,
families satisfying ∀Z ∈ Ob C
Q Q
Hom(U, Z) / Hom(Ui , Z) // Hom(Ui ×U Uj , Z)
is exact, and similarly for a base extension {Ui ×U V → V }. (One has to check
the axioms).
TAUTOLOGY. Every representable functor on C is a sheaf of sets on TC .

It is useful to know that the coverings in a topology are universal effective epi-
morphisms, preferably in some large category, so that one can lay one’s hands on
some sheaves. For instance, in example 1.1.1, the coverings are universal effective
epimorphisms in the category of all topological spaces, i.e., HomT op ( , Y ) is a
sheaf on T for Y a topological space.
Example 1.1.3. Let T = TSets . Then the coverings {Ui → U } are families of
maps which are surjective (i.e. such that U is covered by the union of the images
of the Ui ’s). One verifies easily that every sheaf of sets on T is a representable
functor, namelyF(U ) ≈ Hom(U, F(e)) where e is a set of one element. More
generally
Example 1.1.4. Let G be a group and TG the canonical topology on the category
of left G-sets (sets with G operating). Again coverings are families of G-maps
which are surjective, and every sheaf of sets is a representable functor, in fact
F(U ) ≈ HomG (U, F(G)) where F(G) is obtained by viewing G as a left G-set and
has as operation of σ ∈ G the one induced by right multiplication on G by σ −1 .
2
1 PRELIMINARIES
Therefore the category of abelian sheaves on TG is equivalent with the cate-

gory of G-modules, and so the derived functors of a suitable left exact functor Γ:
ab. sheaves → Ab will be the ordinary cohomology of groups with values in the
corresponding G-module. The functor Γ is of course F Ã F(e) where e is a set of
one element with its unique structure of G-set.
Example 1.1.5. Let G be a profinite group and set TC = category of finite

sets with continuous G-operation; Cov TG = finite families of maps which are
surjective. One may verify that the category of abelian sheaves on TG is equivalent
with the category of continuous G-modules (although this is not the canonical
topology on TG , and only the ”finite” sheaves are representable). Hence, taking
derived functors of Γ : F Ã F(e) one obtains the Tate cohomology groups.
Example 1.1.6. Let X be a nötherian scheme, and define TX by
TX = category of schemes Y/X étale, finite type
Cov TX = finite surjective families of maps
These coverings are universal effective epimorphisms in the category of all preschemes.
For instance, if X = spec k, k a field, then TX is dual to the category of

finite separable (commutative) k-algebras and by Galois theory is equivalent to
the category of finite sets with continuous G(k̄/k) operation, where G(k̄/k) is the
Galois group of the separable algebraic closure k̄ of k. As in example 1.1.5 one
gets Galois cohomology. This example will be examined in more detail later.
1.2 Generalities on lim

−→
Let I, C be categories and F : I → C a functor. For X ∈ C denote by
cX : I → C the constant functor carrying Ob I to X and F u I to idX . We obtain
a covariant functor Homf un (F, cX ) : C → Sets which is denoted by lim F . If
−→
this functor is representable the object representing it is called lim F . Dually,
−→
Hom(cX , F ) is denoted by lim F .
←−
Let us write Fi for F (i), i ∈ Ob I.
3
1.2 GENERALITIES ON lim
−→
Proposition 1.2.1. If C = Sets (resp. Ab) then lim F is representable.

−→
Proof : Take
a M
lim F = Fi /R ( resp. Fi /R)
−→
i∈I i∈I
where R is the equivalence relation (resp. subgroup) generated by pairs (x, y)

(resp. elements x − y), say x ∈ Fi , y ∈ Fj , such that ∃φ : i → j in I with
[F (φ)](x) = y.
Suppose C = Sets then R can be described as follows: For x ∈ Fi , y ∈ Fi0
we have (x, y) ∈ R iff ∃i = i0 , i1 , · · · , in = i0 and j1 , · · · , jn and xν ∈ Fiν (x0 =
x, xn = y); zν ∈ Fjν and a diagram
j1 _> · · · jnbD (1.2.1)

¡? >> zz
< DD
¡¡¡ >> zzz DD
¡ >> zz DD
¡¡ zz D
i0 i1 in
in I with
: z1 ·= · · dII
vvv `BBBB {{{ II
vv BB {{ II
vv BB { II
vv {{ I
x = x0 x1 xn = y
under the induced maps. A diagram 1.2.1 is called a connection of (i, i0 ) in I.
I is connected iff every pair (i, i0 ) of elements has a connection. Obviously if
`
I = Iα is a direct sum (i.e. disjoint union) of categories I then there is a
α
canonical isomorphism
a
lim F |Iα → lim F
−→ −→
α
(this doesn’t depend on C). In particular, if I is a discrete category (a category

in which the arrows are reduced to identity maps) and if direct sums exist in C
then
a
lim F ≈ Fi
−→
i∈I
Q. E. D.
4
1 PRELIMINARIES
The following axioms for I are useful:
L1 Given a diagram
j j=
££A = ==
££ ==
£££ ==
£ Á
i ;; in I, ∃ k
;; ££@
;; ££
;; £££
À £
j0 j0
such that the resulting square commutes.
L2 Given a diagram i ⇒ j, ∃ a map j → k such that the two maps i → k

obtained by composition are the same.
L3 I is connected.
Note that if C = Sets and (L1) holds then the equivalence relation R reads x ∼ y
(say x ∈ Fi , y ∈ Fj ) iff
i ==
==
==
==
Á
∃ k in I
££@
££
£££
£
j
such that the induced images of x, y in Fk are equal. In other words, one may talk
more or less as is usual with inductive systems if (L1) holds in I and C = Sets.
Now let C = Ab and denote by F the category of functors F : I → Ab. F is
an abelian category and
Proposition 1.2.2. The functor F → Ab defined by F Ã lim F is right exact.

−→
We omit the proof.
Let F ∈ F and denote by set −−−lim
−→F the limit of the underlying sets of F .
Then by definition of lim for sets there is a map set
−−−lim
−→F → lim F . In general
−→ −→
this map is not bijective. However,
5
−→
Proposition 1.2.3. Suppose I 6= ∅ and that (L1,2,3) hold in I. Then set

−−−lim
−→F →
lim F is bijective.
−→
Proof: Suppose first (L1,3) hold, and let
n
X M
xν ∈ Fi
ν=1 i∈I
where say xν ∈ Fiν . Applying (L3) and (L1) we can find a diagram
i1 =
==
==
==
=Á
iν /j
¢ @
¢¢¢
¢¢
¢¢
in
P
in I. Let zν ∈ Fj be the image of xν under the induced map, and set z = zν .
P
Then xν ≡ z( mod R) where R is the subgroup defined in the proof of 1.2.1.
This shows that the map
a
Fi → lim F
−→
i∈I
is surjective, and hence set

−−−lim
−→F → lim F is obviously surjective.
−→
Now suppose also (L2) holds. We shall define a group law on set
−−−lim
−→F so
that the maps Fi → −set
−−lim
−→F are homomorphisms. This will induce a map
lim F → set
−−−lim
−→F which, composed with the one under consideration, gives the
−→
identity, proving the proposition.
Let x̄, ȳ ∈ set

−−−lim
−→F be represented by x,y. We may assume x, y ∈ Fi , i ∈ I.
Set x̄ + ȳ = x + y. The group axioms will be trivial if this is well defined, as
will be the fact that Fi → set −−−lim
−→F is a homomorphism. So suppose x̄, ȳ are also
represented by x0 , y 0 ∈ Fi0 .
6
1 PRELIMINARIES
Then there exist diagrams
i << i ==
<< ==
<< ==
<< ==
Á Á
j ; j 0
££A ¢¢@
££ ¢
£££ ¢¢¢
£ ¢
i0 i0
in I with x = x0 in Fj ; y = y 0 in Fj 0 . Apply (L1) to the diagram
Aj
¤¤¤
¤
¤¤
¤¤
i @@
@@
@@
@@
Â
to get
@j >>
£££ >>
£ >>
££ >>
££ Á
i << @¢ k
<< ¢
<<
<< ¢¢¢
Á ¢¢
j0
which commutes. We get two maps i0 → k. Apply (L2) to find k → k such that
the two maps i0 → k 0 are equal. Then
@j> @j ==
£££ >>> ¢¢¢ ==
£ >> ¢ ==
££ >> ¢¢ ==
££ Â ¢¢ Á
i << k0 and i0 == @k
<< ¡¡? == ¢¢
<< ¡¡ == ¢¢
<< ¡¡ == ¢¢
À ¡ ¡ Á ¢¢
j0 j0
both commute and so the uniquely determined images in Fk0 of x, x’(resp. y, y’)
are equal. Hence x + y = x0 + y 0 .
Q. E. D.
7
−→
Corollary 1.2.4. If (L1,2) hold in I then lim is an exact functor: F → Ab.

−→
0 0 0
It is clear that set
−−−lim
−→F ,→ set
−−−lim
−→F if F ,→ F (i.e. Fi ,→ Fi for each i).
Hence we are done if also (L3) holds by 1.2.2 and 1.2.3. If (L3) does not hold
` L
write I = Iα with Iα connected and use the fact that is an exact functor.
α
Definition 1.2.1. A subcategory J ⊂ I is a final subcategory (formerly: cofinal)
iff
i J is a full subcategory.
ii ∀i ∈ I, ∃i → j in I with j ∈ J .
One verified easily the following:

Proposition 1.2.5. Let J ⊂ I be a final subcategory. (Lx) for I ⇒ (Lx) for J
(x=1,2,3). If I satisfies (L1) and F : I → C is arbitrary then the canonical map
lim F → lim F |J is bijective. In particular, if I has a final object ∞ (equivalently:
−→ −→
{∞, id∞ } is a final subcategory of I and (L1) holds in I) then lim F ≈ F∞ .
−→
Let again I be given, and a ∈ I. Denote by a\I the category of maps a → i
of I and if F : I → C is a functor, denote by a\F (a → i) = F (i).
Proposition 1.2.6. Let I, a, F be as above. Then (L1) for I ⇒ (L1,3) for a\I
and (L2) for I ⇒ (L2) for a\I. If I satisfies (L1,2,3) then the canonical map
lim F → lim a\F is bijective.
−→ −→
We verify the last assertion: Let X ∈ C. We have to show that the obvious
map Hom(F, cX ) → Hom(a\F, a\cX ) is bijective. This means we have to show
that an element ξ ∈ Hom(a\F, a\cX ) (i.e. a collection of maps Fi → X for each
a → i ∈ a\I with the appropriate commuting relations) determines a unique
element of Hom(F, cX ). So let ξ be given and try to define for i ∈ I a map
Fi → X. Applying (L3), we find a diagram
i <<
<<
<<
<<
Á
@¢ j in I
¢¢
¢¢¢
¢¢
a
8
1 PRELIMINARIES
Hence ξ includes a map Fj → X and we are forced to define Fi → X by com-

position Fi → Fj → X. The commuting relations will be obvious if this is well
defined. Suppose also
i ==
==
==
==
Á
Ä?
i0 in I
Ä
ÄÄ
ÄÄÄ
Ä
a
Aj
¤¤¤
¤
¤¤¤
¤
Apply (L1) to the diagram i / j0 to find
@j >>
£££ >>
£ >>
£££ >>
£ Á
i << @k
<<
<< ¢¢¢
<< ¢¢
Á ¢¢
j0
which commutes. We get two maps a → k, and applying (L2) we may assume
they are equal. Then ξ includes a uniquely determined map Fk → X and since ξ
is a morphism of functors, Fj → X (resp. Fj 0 → X) is obtained by composition
with Fj → Fk (resp. Fj0 → Fk ). Therefore the maps Fi → X defined above are
the same, namely they are obtained by Fi → Fk → X. Done.
Proposition 1.2.7. Let F, G, H : I → Sets be three functors and let F → G,

H → G be morphisms. Suppose I satisfies (L1,2,3). Then
∼
lim(F ×G H) −→ lim F ×lim G lim H
−→ −→ → −→
−
MORAL. Axioms (L1,2,3) make I as good as an inductive system for limits.
Proposition 1.2.8. Let I be given, F the category of functors I → Ab. For

F ∈ F, lim F is representable and the functor F Ã lim F is left exact.
−→ −→
9
1.3 PRESHEAVES
1.3 Presheaves
Let C be a category and denote by P = PC the category of functors F :
C ∗ → Ab. We refer to such functors as (abelian) presheaves on C. (The reader
may, inserting axioms where necessary, replace Ab by an arbitrary category if he
feels so inclined.) All of the usual constructions available in Abmay be made in
P (by doing them for each U ∈ C) and all functorial properties are preserved. In
particular, P is an abelian category and satisfies axioms AB6,4(cf. [4] Proposition
1.6.1). A sequence F 0 → F → F 00 ∈ P is exact iff F 0 (U ) → F (U ) → F 00 (U ) is
exact for each U ∈ C.
Let C 0 be another category and f : C → C 0 a functor. Define a functor
f P : PC 0 = P 0 → P by f P (F ) = F ◦ f . The functor f P is obviously exact.
α
Recall that if M À are categories and functors then β is left adjoint to α
β
iff the functors Hom(Y, αX), Hom(βY, X) : N × M → Sets are isomorphic. (If
α(resp. β) is given then β(resp. α) is unique up to isomorphisms if it exists).
Theorem 1.3.1 (Kan). There is a functor fp : P → P 0 left adjoint to f P is right

exact.
We sketch the proof for the convenience of the reader:

For Y ∈ C 0 , let IY = IYf be the following category:
Ob IYf = {pairs(X, φ)|X ∈ ob C, φ ∈ HomC 0 (Y, f (X))}

with (1.3.1)
Hom((X1 , φ1 ), (X2 , φ2 )) = {ξ ∈ Hom(X1 , X2 )|f (ξ)φ1 = φ2 }
Notice
²
(a) If Y → Z in C 0 we get a functor ²̄ : IY → IZ in the obvious way.
(b) If F is a presheaf on C, we get a functor FY : IY◦ → Ab by FY ((X, φ)) =

F (X). It depends ”functorially” on F.
Now set
fP F (Y ) = lim FY = lim F (X)
−→ −→
(X,φ)∈IY
10
1 PRELIMINARIES
fP F is made into a presheaf on C 0 (i.e. a functor) by (a). Because of (b), F Ã fP F

is a functor, right exact by Proposition 1.2.2. To show fP is adjoint to fP we
∼
have to define for F ∈ P, G ∈ P 0 an isomorphism Hom(F, f P G) → Hom(fP F, G)
which commutes with maps in P, P 0 .
Suppose ξ ∈ Hom(F, f P G). Then we have for each X ∈ C a map F (X) →
G(φ)
G(f (X)). Hence for Y ∈ C 0 , (X, φ) ∈ IY we get a map F (X) → G(f (X)) −→
G(Y ), and therefore a map lim FY = fP F (Y ) → G(Y ), whence fP F → G.
−→
Conversely, if η ∈ Hom(fP F, G); i.e., for Y ∈ C 0 , fP F (Y ) → G(Y ) then in
particular if Y = f (X) we have fP F (f (X)) → G(f (X)). Now (X, idf (X) ) ∈ If (X)
and so there is a map F (X) → fP F (f (X)). Composing, we find F (X) →
G(f (X)), whence F → f P G. These maps are obviously inverses of each other
and so the theorem follows (of course, some details were omitted).
We want also to recall the following:
Corollary 1.3.2. If fP is exact then f P carries injectives into injectives.
For, if G ∈ P 0 is injective, i.e., Hom(x, G) is an exact functor for x ∈ P 0 we

want to know Hom(x, f P G) exact for x ∈ P. Since fP is exact and G is injective
we are done.
Example 1.3.1. Let X ∈ C and denote by {X} the discrete category Ob {X} =
{X}, F u {X} = {idX }. Let i :→ {X} → C be the inclusion. A presheaf on {X}
is just an abelian group, and for F ∈ PC , iP F = F (X). We claim iP is exact,
and to show it we need to show lim is exact for the categories IYi , Y ∈ C defined
−→
in 1.3.1. But clearly for (X, φ), (X, ψ) ∈ IYi , Hom((X, φ), (X, ψ)) = ∅ unless
φ = ψ and = {idX } if φ = ψ. In other words IYi is the discrete category on the
L
set Hom(Y, X). So if A is an abelian group then iP A(Y ) = A. This is
Hom(Y,X)
certainly exact.
Corollary 1.3.3. If F ∈ PC is injective then F (X) is an injective abelian group

for each X ∈ C.
For, apply 1.3.2.
11
1.4 ČECH COHOMOLOGY
Notation 1.3.1. Let i : {X} → C be as above. We write ZX = iP Z, so ZX ∈ PC

and we have canonically, for F ∈ PC 0 , Hom(ZX , F ) ≈ Hom(Z, F (X)) ≈ F (X),
i.e., ZX represents the (covariant) functor P → Ab given by F Ã F (X). ZX
depends covariantly on X ∈ C. Note that the presheaves ZX , X ∈ C, are a set of
generators for PC . Hence (cf. [4], Theorem 1.10.1)
Corollary 1.3.4. Every F ∈ PC can be embedded in an injective.

f
Example 1.3.2. Suppose C À C 0 are given with g left adjoint to f and consider
g
the category IYf , 0
Y ∈ C . Recall
Ob IYf = {(X, φ)|X ∈ C, φ : f (X) ← Y }
By adjointness of g to f this is the same as {(X, ψ)|X ∈ C, ψ : X ← g(Y )}.

Therefore it is clear that IYf has a final object, namely (g(Y),id), and so if F ∈ PC ,
fP F (Y ) ≈ F (g(Y )) = g P F (Y ). Hence fP ≈ g P and so f P carries injectives into
injectives. This situation arises for instance if one tries to relate C to C/Y (Y ∈ C
fixed), and if X × Y exists in C, ∀X.
1.4 Čech cohomology

Let C be a category and {Uα → V }α∈I a family of maps in C. Suppose the
products used below exist in C. We get a lot of maps denoted symbolically by
the figure below
o 1̂
V o {Uα }α∈I o 0̂ {Uα ×V Uβ }(α,β)∈I 2
o 2̂
o 1̂
{Uα ×V Uβ ×V Uγ }(α,β,γ)∈I 3 · · ·
o 0̂
If F is a presheaf on C we obtain a diagram

F (0̂)
Q /Q
α F (Uα ) F (1̂)
/ α,β F (Uα ×V Uβ )
F (0̂)
/
F (1̂)
/ {Uα ×V Uβ ×V Uγ }(α,β,γ)∈I 3 · · ·
F (2̂)
/
12
1 PRELIMINARIES
in the canonical way. Define

Y Y
dn : F (Uα0 ×V · · · ×V Uαn ) → F (Uα0 ×V · · · ×V Uαn+1 )
α0 ,··· ,αn α0 ,··· ,αn+1
by
n+1
X
dn = (−1)i F (î)
i=0
and verify dn+1 ◦ dn = 0. One obtains a functor PC →(cochain complexes) which

is obviously exact (since AB4* holds in Ab). Taking cohomology, we get an exact
sequence of cohomological functors PC → Ab denoted by H q ({Uα → V }, ), where
Y Y
H 0 ({Uα → V }, F ) = ker( F (Uα ) ⇒ F (Uα ×V Uβ ))
α α,β
These functors are called the Čech cohomology of the family {Uα → V }.
Theorem 1.4.1. The functors H q (q > 0) defined above are effacable functors.
Corollary 1.4.2. They are the derived functors of the left exact functor H 0 .
Proof of Theorem 1.4.1: Suppose F ∈ PC is injective and let ZX ∈ PC be as in

1.3.1. We want to show the cochain complex
Q d1 Q d2
F (Uα ) / F (Uα ×V Uβ ) / ···
defined above is exact. This is the same as showing

Q Q
Hom(ZUα , F ) / Hom(ZUα ×V Uβ , F ) / ···
= Hom(⊕ZUα , F ) / Hom(⊕ZU × U , F ) / ···

α V β
exact.
Since F is injective it suffices to show
⊕ZUα o ⊕ZUα ×V Uβ o ···
is an exact sequence of presheaves (where the maps are determined by the previous
line), i.e., to show that for each Y ∈ C
⊕ZUα (Y ) o ⊕ZUα ×V Uβ (Y ) o ··· (1.4.1)
13
is exact. Remembering that ZX (Y ) = ⊕Hom(Y,X) Z one sees easily that the se-
quence 1.4.1 is induced by the obvious maps in the diagram
` o `
Hom(Y, Uα ) o Hom(Y, Uα ×V Uβ ) · · ·
α (α,β)
` ` `
= [ Homφ (Y, Uα ) oo Homφ (Y, Uα ×V Uβ ) · · · ]
φ∈Hom(Y,V ) α (α,β)
Set a
S(φ) = Homφ (Y, Uα )
α
a
= [ S(φ) oo S(φ) × S(φ) · · · ]
φ∈Hom(Y,V )
Hence we need to show that a diagram of the form

L o L o L
Zo Z oo Z···
S S×S S×S×S
is exact. But the sequence is homotopically trivial, the homotopy being given by
n(i0 , · · · , in ) Ã n(l, i0 , · · · , in )(l ∈ S fixed)
for n(i0 , · · · , in ) = n in the (i0 , · · · , in )-th component of ⊕S n+1 Z. This completes

the proof.
Q. E. D.
Definition 1.4.1. Let
{Uα → V }α∈I , {Uα0 → V }α∈I 0

f ²
be families of maps in C. A map {Uα → V } → {Uν0 → V } is a map I → I 0 and
fα
for α ∈ I a map Uα → U²(α) such that
Uα @ / U²(α)
@@ {{
@@ {{
@@ {{
@Ã {} {
V
commutes.
14
1 PRELIMINARIES
One obtain a diagram
V o {Uα } oo {Uα ×V Uβ } · · ·
f f ×f
² ²
V o {Uν0 } oo {Uν0 ×V Uµ0 } · · ·
where everything commutes, and hence a map of f ∗ of complexes

Q Q
F (Uα ) / F (Uα ×V Uβ ) / ···
O O
f1∗ =F (f ) f2∗ =F (f ×f )
Q Q
F (Uν0 ) / F (Uν0 ×V Uµ0 ) / ···
which induces a map on Čech cohomology. It is well known that:
Proposition 1.4.3. If f, g : {Uα → V }{Uν0 → V } are any two maps then f ∗ , g ∗

are homotopic, and hence the induced maps on cohomology are equal.
We omit the proof. To get the last fact one has anyhow only to check it for
H 0 , by Corollary 1.4.2, and this is easy.
15
16
2 SHEAVES
2 Sheaves
2.1 Sheaves
Let T be a topology (Definition 1.1.1), let P be the category of abelian
presheaves on T and S ⊂ P the set of sheaves. We make S into a full subcategory
of P, i.e., we define a morphism of sheaves to be a morphism of presheaves. Denote
by i : S ,→ P the inclusion.
Theorem 2.1.1. There is a functor ] : P → S which is adjoint to i.
If P ∈ P, P ] = ](P ) is called the associated sheaf to P. Since S is a

subcategory of P the adjointness property can be stated as follows: There is a
functorial homomorphism P → ](P ) which has the universal mapping property
for maps of P into sheaves. The proof is given in the following pages. It can be
used without change also for presheaves and sheaves with value in Sets.
We begin by defining a functor + : P → P: Let U ∈ T be fixed and denote
by JU the category of coverings {Uα → U } of U in Cov T. with maps in the sense
of 1.4.1. A presheaf P ∈ P induces a functor PU : JU◦ → Ab by
Q
PU ({Uα → U }) = ker( P (Uα ) // P (Uα ×U Uβ ))
= H 0 ({Uα → U }, P )
φ J(φ)
Note that for V → U we get also a functor JU −− −→ JV by {Uα → U } Ã
{Uα ×U V → V }(cf. 1.1.1,3), and a morphism of functors PU → PU ◦ J(φ) since
Uα ×U V o Uα ×U Uβ ×U V ≈ Uα ×U V ×V Uβ ×U V
² ²
Uα o Uα ×U Uβ
commutes and since ker is a functor. Therefore we get lim PU → lim PV .

−→ −→
+ 0 +
Set P (U ) = lim PU (, H (T, U ; P )) and give P the structure of presheaf
−→
just defined.
17
2.1 SHEAVES
Remark 1. JU is not a good category for limits, but because of 1.4.3 the ho-
momorphism PU ({Uα → U }) → PU ({Vν → U }) is uniquely determined if there
exist maps {Vν → U } → {Uα → U }, and so PU : JU◦ → Ab can be factored
through JU → J¯U where J¯U is the partially ordered set obtained by writing
{Vν →≥ {Uα → U }} iff ∃ a map {Vν → U } → {Uα → U }. J¯U is an induc-
tive system because (L3) (cf. Section 1.1) holds in JU . In fact, if {Uα → U },
{Vν → U } ∈ Cov T then {Uα ×U Vν → U } ∈ Cov T by axioms 3, 2 of 1.1.1, and
we have canonical maps {Uα → U } ← {Uα ×U Vν → U } → {Vν → U }
Q
It follows, since is exact and ker is left exact, that + : P → P is left exact.
Remark 2. For every {Uα → U } ∈ JU there is a canonical map P (U ) →

Q
PU ({Uα → U }) given by P (U ) → P (Uα ). Since (L3) holds in JU the limit
of the constant functor JU◦ → Ab with value P (U ) is canonically isomorphic to
P (U ). We get a functorial homomorphism P → P + and by definition of sheaf, if
S ∈ S then S → S + is bijective. Hence any map of P into a sheaf must factor
through P + . Therefore Theorem 2.1.1 will be proved if we show
SURPRISE. P ++ is a sheaf
Definition 2.1.1. Let P ∈ P, P satisfies (+) iff

Y
∀{Uα → U } ∈ Cov T, P (U ) ,→ P (Uα )
”SURPRISE” (and hence 2.1.1) follows from the following
Lemma 2.1.2. (i) P + satisfies (+). (ii) If P satisfies (+) then P + is a sheaf.
Proof of (i): Let P ∈ P, {Uα → U } ∈ Cov T be given. Let ξ¯1 , ξ¯2 ∈ P + (U ) and
suppose ξ¯1 , ξ¯2 have the same image in P + (Uα ), all α. ξ¯1 , ξ¯2 may be represented
by elements
Q Q
ξ1 , ξ2 ∈ ker( P (Vν )) // P (Vν ×U Vν 0 )
for some {Vν → U } ∈ Cov T (we are tacitly using remark 1). Now the image of
ξ¯i in P + (Uα ) is then represented by the image of ξi in
Q / Q
ker( P (Uα ×U Vν ) / P (Uα ×U Vν ×U Vν 0 ))
ν ν,ν 0
18
2 SHEAVES
Since ξ¯1 = ξ¯2 there is a ”finer” covering {Wαµ → Uα } such that the images of ξ1 , ξ2
Q
in P (Wαµ ) are equal. But then, letting α vary, the family {Wαµ → U } ∈ Cov T
µ
Q
is a refinement of {Vν → U } and ξ1 = ξ2 in α,µ P (Wαµ ), so ξ¯1 = ξ¯2 .
Proof of (ii): Note first that
f
Claim 2.1.3. If P satisfies (+) and if {Vν → U } −→ {Uα → U } in JU then
Q Q Q Q
the map ker( P (Uα ) ⇒ P (Uα ×U Uβ )) → ker( P (Vν ) ⇒ P (Vν ×U Vµ )) is
injective.
For consider the diagram
{Vν ×U Uα }
rr MMM
MMMp2
p1
rrrrr MMM
rx rr f
M&
{Vν } M / {Uα }
MMM q
MMM qq
MMM qqqqq
MM& q
qx qq
U
Holding α fixed, {Vν ×U Uα → Uα } ∈ Cov T by axiom 3 of 1.1.1. Hence,

Q ÂÄ /
Q
P (Uα ) P (Vν ×U Uα ) because of (+). By axiom 2 for T, {V × U →
α,ν ν U α
α
U }α,ν ∈ Cov T . Combining, we find the map
Q /Q Q /
ker( P (Uα ) / P (Uα ×U Uα0 )) / ker( P (Vν ×U Uα ) /)
induced by p2 is injective. But this map is unique (cf. 1.4.3) and is therefore the
same as the one induced by p1 ◦ f . So the one under consideration (induced by
f) is also injective.
Now say P satisfies (+), let {Vα → U } ∈ Cov T and
Q /Q
ξ¯ ∈ ker( P + (Vα ) / P + (Vα ×U Vβ ))
We have to show ξ¯ is image of some element in P + (U ). Choose for each α a family

Q /
{Wαν → Vα } ∈ Cov T with ξα ∈ ker( ν P (Wαν ) / ) representing the α-th
19
2.1 SHEAVES
component ξ¯α of ξ.
¯ Consider the diagram (in which all squares are cartesian)
{Wαν } o {Wαν ×U Vβ } o {Wαν × Wβµ }
² ² ²
{Vα } o {Vα ×U Vβ } o {Vα ×U Wβµ }
² ² ²
Uo {Vβ } o {Wβµ }
We find ξα induced by base extension an element

Q Q
1
ξαβ ∈ ker( P (Wαν ×U Vβ ) / P (Wαν ×U Vβ ×Vα ×U Vβ Wαν 0 ×U Vβ ))
/
ν ν,ν 0
and ξβ induces an element

Q Q
2
ξαβ ∈ ker( P (Vα ×U Wβµ ) / P (Vα ×U Wβµ ×Vα ×U Vβ Vα ×U Wβµ0 ))
/
µ µ,µ0
¯ ξ 1 and ξ 2 represent the same element of P + (Vα ×U Vβ ).

By assumption on ξ, αβ αβ
2
Hence ”ξα1 = ξαβ ” in some covering of Vα × Vβ which is a common refinement of
{Wαν ×U Vβ → Vα ×U Vβ }ν and {Vα ×U Wβµ → Vα ×U Vβ }µ . By 2.1.3 this must
Q
be so in any common refinement, and hence ”ξα1 = ξαβ
2
” in P (Wαν ×U Wβµ ).
ν,µ
This shows Q Q
ξ ∈ ker( P (Wαν ) / P (Wαν ×U Wβµ ))
/
α,ν α,ν,β,µ
whence ξ¯ ∈ P + (U )
Q. E. D.
Having theorem 2.1.1 one can copy large parts of Godement’s book. In
particular,
Theorem 2.1.4. .
i S is an abelian category satisfying AB5, AB3* and has generators.
ii i : S ,→ P is left exact and ] : P → S is exact.
20
2 SHEAVES
Proof of (i): Let F → G ∈ S, and set K=”Presheaf cokernel”. Then if {Uα →

U } ∈ Cov T we have
0 0 0
² ² ²
/
Q /Q
K(U ) K(Uα ) / K(Uα ×U Uβ )
² Q ² ²
/ F (U ) / /Q
0 F (Uα ) / F (Uα ×U Uβ )
² Q ² ²
/ G(U ) / /Q
0 G(Uα ) / G(Uα ×U Uβ )
Since ker is a left exact functor in Ab it follows that
Q
/ ker( K(Uα ) /Q
K(U ) / K(Uα ×U Uβ ))
i.e., that K is a sheaf. Obviously, therefore, K is a kernel in S (i.e., 0 →

Hom(X, K) → Hom(X, F ) → Hom(X, G) is exact, X ∈ S).
Let C=”presheaf coker(F → G)”. Then for X ∈ S we have (by the universal
mapping property of P → P ] )
Hom(F, X) o Hom(G, X) o Hom(C, X) o 0

∼
= ∼
= ∼
=
² ² ²
Hom(F ] , X) o Hom(G] , X) o Hom(C ] , X) o 0
Here the top line is exact by definition of C, and since F, G are sheaves, F ≈ F ]
and G ≈ G] . Therefore C ] (with the induced map G → C ] ) is a cokernel in S
Finally, let I=”presheaf coker(K → F )”=”presheaf coimage (F → G)”.
Then I ] = ”sheaf coimage (F → G)” with the obvious map. We need to show I ]
is the image = ker(G → C ] ). Now certainly 0 → I → G → C is exact in P since
P is an abelian category. Therefore, since i ◦ ] : P → P is left exact because + is
(cf. remark 1), 0 → I ] → G] → C ] is exact in P. Recalling the discussion of ker
above and that G −−∼−→ G] we are done.
AB3* (existence of product): Let Fi , i ∈ I be sheaves. It is immediate that
the presheaf product of the Fi s is a sheaf, and hence a fortiori a product in the
category of sheaves.
21
2.1 SHEAVES
AB5 (existence of sums, etc.): Let Fi , i ∈ I be sheaves, and F=”presheaf

sum” of the Fi s. By the universal mapping property of ], it is clear that F ] is a
sum in the category of sheaves.
Let C, B be sheaves, and Ai ⊂ C(i ∈ I) an increasing filtering family of
subsheaves. Suppose given morphisms Ai → B compatible with the inclusions of
Ai in Aj (i ≤ j). To complete the proof of AB5 we have to find a map A → B
inducing the maps Ai → B, where A = sup Ai ⊂ C. Set Ā(U ) = ∪i Ai (U ),
whence Ā ⊂ C is a presheaf. Since C is a sheaf, the inclusion of Ā can be
factored through Ā → Ā] → C, and since ] is left exact, Ā] ,→ C. Obviously
Ā] (with this injection) is canonically isomorphic to A. Since the maps Ai → B
define a map Ā → B and since B is a sheaf, we find the desired map Ā] → B.
Generators: Clearly ZU] is a set of generators for S (cf. 1.3.1)
Notation 2.1.1. We write ZU] = ZU . ZU represents the (covariant) functor

F Ã F (U ) on S.
Proof of (ii): i : S → P is left exact because the presheaf kernel of a map of

sheaves is already a sheaf. ] : P → S is left exact because i ◦ ] is (as was noted
above) and because i is left exact and fully faithful. To show ] right exact, let
P 0 → P → P 00 → 0 ∈ P be exact, and X ∈ S. Then
Hom(P 0 , X) o Hom(P, X) o Hom(P 00 , X) o 0
Hom(P 0] , X) o Hom(P ] , X) o Hom(P 00] , X) o 0
commutes, and the top line is exact, hence the bottom is, and P 0] → P ] → P 00] →
0 is exact.
Q. E. D.
Miscellany 2.1.1. :
i The category S has enough injectives.
ii An injective in S is injective as presheaf.
22
2 SHEAVES
iii For U ∈ T , the functor ΓU : S → Ab given by ΓU (F ) = F (U ) is left exact.
iv Let 0 → F → G ∈ S be exact, and C=”presheaf cokernel (F → G)”. Then

C satisfies (+).
Proof :
i cf. [4], Theorem 1.10.1. ii Because ] is exact (cf. 1.3.2). iii ΓU is the
composition of the left exact functor i and the exact section functor on P. iv We
have a diagram, for {Ui → U } ∈ Cov T :
0 0
² Q ²
/ F (U ) / /
0 F (Ui ) /
² Q ²
/ G(U ) / /
0 G(Ui ) /
² Q ² /
0 / ker / C(Ui ) /
²
0
Q
Hence C(U ) = G(U )/F (U ) ,→ ker( C(Ui ) ⇒).
Q. E. D.
HOMEWORK. Discuss Hom(F, G).
2.2 Cohomology
Let f : S → A be any left exact functor, where S is as above and A is an
abelian category. Because of 2.1.1,i, one can define the right derived functors Rq f
of f (cf. [4], Section 2.3).
Notation 2.2.1. Let U ∈ T . The derived functors of Γ : S → Ab (cf. 2.1.1,iii)

are denoted by Rq f ΓU = H q (T, U ; ). So in particular, for F ∈ S, H 0 (T, U ; F ) =
F (U ).
23
2.2 COHOMOLOGY
Notation 2.2.2. Fixing F ∈ S, we get a functor Γ(F ) : T ◦ → Ab by U Ã

F (U ) = Γ(F ). Denote by ←Γ
−: S → Ab the functor F Ã lim Γ(F ) = lim F (U ).
←− ←−
U ∈T ◦
Because of Proposition 1.2.8, it is clear that ←
Γ
− is a left exact functor. The right
derived functors are denoted by Rq ← Γ q
− = H (T, ).
Clearly, if T has a final object X (initial in T ◦ ), then H q (T, F ) ≈ H q (T, X; F ),

F ∈ S.
Notation 2.2.3. The right derived functors of i : S → P are denoted by Rq i =

Hq ( ).
So for F ∈ S the Hq (F ) are presheaves, and H0 (F ) = F viewed as a presheaf.

In fact, we have canonically
[Hq (F )](U ) ≈ H q (T, U ; F ) (2.2.1)
To show this we remark that H q (T, ; F ) is a functor of U (i.e., a presheaf)

since F is. To show H q (T, ; F ) are the derived functors of i we have only to check
a They agree if q = 0.
b They vanish on injectives for q > 0.
c They form an exact cohomological functor.
All of these assertions are trivial.
Proposition 2.2.1. (Hq (F ))+ = 0 for F ∈ S, q > 0.
Proof : It follows from Lemma 2.1.2 that for P ∈ P, P + ,→ P ] . Hence we need

only show (Hq (F ))] = 0. Now ] ◦ i ≈ idS . The functor ] is exact (by 2.1.4) and so
its derived functors vanish. We get a spectral sequence for the composed functor
(cf. [4], Theorem (2.4.1)) from which one can read off the desired result.
Q. E. D.
24
2 SHEAVES
2.3 The spectral sequences for Čech cohomology

Let {Uα → U } ∈ Cov T , and F ∈ S. Then F (U ) is functorially isomorphic
to H 0 ({Uα → U }, F ) (cf. Section 1.4), by definition of sheaf. In other words,
the functor ΓU defined above is isomorphic to the composition of the functors
i, H 0 ({Uα → U }, ). Since i carries injectives into injectives (2.1.1,iii) we obtain
(cf. [4], Theorem 2.4.1)
Spectral Sequence 2.3.1.
E2p,q = H p ({Uα → U }, Hq (F )) =⇒ H ∗ (T, U ; F ) (F ∈ S)
More generally, let U ∈ T be fixed, and J ⊂ JU be any subcategory satisfying

(L3), where JU is as in Section 1. Denote by H 0 (J; ) : P → Ab the functor
H 0 (J; P ) = lim H 0 ({Uα → U }; P )

−→
{Uα →U }∈J
By reasoning similarly to that of remark 1 of Section 1, one sees that H 0 (J; ) is

a left exact functor and that its right derived functors H q (J; ) are obtained as
H q (J; P ) = lim H q ({Uα → U }; P ) (2.3.1)

−→
{Uα →U }∈J
(One uses Corollary 1.4.2 and Proposition 1.4.3 of Chapter 1). Moreover ΓU is
isomorphic to the composition of the functors i and H 0 (J; ). Hence we obtain
for F ∈ S
E2p,q = H p (J; Hq (F )) =⇒ H ∗ (T, U ; F ) (F ∈ S)
In the special case J = JU is the category of all coverings of U, we write
Notation 2.3.1.
H̆ 0 (T, U ; P ) = H 0 (JU ; P ) = P + (U )
H̆ q (T, U ; P ) = H q (JU ; P ) (P ∈ P)
25
2.4 THE CATEGORY T OP , LERAY SPECTRAL SEQUENCES
H̆ q (T, U ; P ) is called the Čech cohomology of P with respect to U. Note that

although it is a functor on P (which depends only on T ), the Čech cohomology
depends also on Cov T. 2.3.2 now reads, for F ∈ S
E2p,q = H̆ p (T, U ; Hq (F )) =⇒ H ∗ (T, U ; F ) (F ∈ S)
Since H̆ 0 (T, U ; P ) = P + (U ), it follows from Proposition 2.2.1 that E2p,q =

0, q > 0. Therefore one reads from 2.3.3
Corollary 2.3.1.
H̆ 1 (T, U ; F ) ≈ H 1 (T, U ; F )
H̆ 2 (T, U ; F ) ,→ H 2 (T, U ; F ) (F ∈ S)
2.4 The category T op, Leray spectral sequences

Recall
Lemma 2.4.1 (Tohoku). Let f : A → A0 be a left exact functor with A, A0

abelian categories and suppose A has enough injectives. If M ⊂ ObA satisfies
a ∀F ∈ A, ∃ 0 → F → M exact in A with M ∈ M.
b F ⊕ G ∈ M =⇒ F ∈ M.
c If 0 → M 0 → M → M 00 → 0 exact in A, M 0 , M ∈ M then F ∈ M and

f (0 → M 0 → M → M 00 → 0) is exact.
then all injectives are in M and Rq f (M ) = 0, all M ∈ M, q > 0. (Hence

resolutions from M may be used to calculate the functors Rq f ).
Definition 2.4.1. Let T be a topology and F a sheaf on T. F is flask iff. for

every {Uα → U } ∈ Cov T , H q ({Uα → U }, F ) = 0, q > 0 (cf. Section 1.4).
Proposition 2.4.2. Let 0 → F 0 → F → F 00 → 0 be an exact sequence of sheaves.

Then
26
2 SHEAVES
i If F’ is flask, the sequence is exact as a sequence of presheaves.
ii F’, F flask =⇒ F” flask.
iii F ⊕ G flask =⇒ F flask.
iv Injectives are flask.
Proof : i: Say F’ flask. To show the sequence exact as a sequence of presheaves it

suffices to show the presheaf coker (F 0 → F ) = C is a sheaf, as then C ≈ C ] ≈ F 00 .
We have to show for {Uα → U } ∈ Cov T that
Q /Q
C(U ) → ker( C(Uα ) / C(Uα ×U Uβ ) ) = H 0 ({Uα → U }, C)
is bijective.
This is obvious from the diagram
0 / H 0 ({Uα → U }, F 0 ) / H 0 ({Uα → U }, F ) / H 0 ({Uα → U }, C) / H 1 ({Uα → U }, F 0 )
O O O
∼
= ∼
= ∼
=
0 / F 0 (U ) / F (U ) / C(U ) /0
1 0
since H ({Uα → U }, F ) = 0 by assumption.
ii: Knowing i, this is clear from the exact cohomology sequences for H q ({Uα →
U }, ).
iii: The cohomology commutes with finite direct sums.
iv: Trivial.
Q. E. D.
Corollary 2.4.3. Flask resolutions can be used to calculate the derived functors
H q (T, U ; ) and Hq ( ) (cf. 2.2.1, 2.2.3).
In fact, applying 2.4.2 to 2.4.1 we find iv =⇒ a, iii =⇒ b and i, iii =⇒ c if

f (F ) = F (U ) or i(F ). It is not true in general that flask resolutions can be used
to calculate H q (T, ) (cf. 2.2.2).
27
Definition 2.4.2. We define the category T op of topologies as follows: ObT op =

set of topologies, and, for T, T 0 ∈ ObT op, Hom(T, T 0 ) = set of functors f : T →
T 0 satisfying:
φα
If {Uα −→ U } ∈ Cov T and V → U ∈ T then
f (φα )
i {f (Uα ) −→ f (U )} ∈ Cov T 0 .
∼
ii f (Uα ×U V ) −→ f (Uα ) ×f (U ) f (V ) for all α (with the canonical map).
A functor f ∈ Hom(T, T 0 ) is called a morphism of topologies.
Example 2.4.1. Let T, T’ be the topologies on topological spaces X, X’ respec-

tively. A continuous map π : X 0 → X induces, as is well known, a morphism
f : T → T 0 . If X ⊂ X 0 is an open subset, then the inclusion T 0 ⊂ T is a
morphism, adjoint in fact to the restriction T → T 0 above.
Let T, T 0 ∈ ObT op. Denote by ST , ST 0 , PT , PT 0 the categories of sheaves and

presheaves on T, T’ and by i, i’, ], ]0 the functors of Section 2.1. Let f : T → T 0
be a morphism. It is immediate from the definition above that if F 0 ∈ ST0 then
f P F 0 ∈ ST (where f P is as defined in Section 1.3). We get a functor
f S : ST 0 → ST ; f S = f P ◦ i0 ≈ ] ◦ f P ◦ i0
which is clearly left exact. So we have for F 0 ∈ ST 0 and U ∈ T , f S F 0 (U ) =

F 0 (f (U )). One verifies readily that f S has as left adjoint
fS : ST → ST 0 ; fS = ]0 ◦ fP ◦ i
Note however: It is not clear that fS is exact, even if fP is.

The right derived functors Rq f S of f S can be described as follows: ] and f P
are exact, hence ] ◦ f P is exact, and so its derived functors vanish. Therefore,
since f S ≈ ] ◦ f P ◦ i0 we have, for F 0 ∈ ST 0 and Hq (F 0 ) = Rq i0 F 0 ,
Rq f S (F 0 ) ≈ [f P (Hq (F 0 ))]] (here p is not an integer)
Or, writingRq F 0 = f p (Hq (F 0 )) and referring to 2.2.1 we get
28
2 SHEAVES
Corollary 2.4.4. Rq f S (F 0 ) ≈ [Rq F 0 ]] where for U ∈ T [Rq F 0 ](U ) = H q (T 0 , f (U ); F 0 ).
Lemma 2.4.5. Let f ∈ Hom(T, T 0 ) and F 0 ∈ ST 0 . If F’ is flask, so is f S F 0 .
Proof :Let {Uα → U } ∈ Cov T . Applying 2.4.2, ii repeatedly, we find f (Uα0 ×U

∼
· · · ×U Uαn ) −→ f (Uα0 ) ×f (U ) · · · ×f (U ) f (Uαn ) (where the map is canonical).
Therefore
Y Y
f S F 0 (Uα0 ×U · · · ×U Uαn ) = F 0 (f (Uα0 ×U · · · ×U Uαn ))
(α0 ,··· ,αn ) (α0 ,··· ,αn )
∼
Y
←− F 0 (f (Uα0 ) ×f (U ) · · · ×f (U ) f (Uαn ))
(α0 ,··· ,αn )
This isomorphism commutes with the projection maps and so the lemma follows
from 2.4.2, i and the definition of flask.
Q. E. D.
g f
Corollary 2.4.6. Let T 00 /T / T 0 be morphisms of topologies, and F 0 ∈
ST 0 a flask sheaf. Then f S F 0 is Rq g S -acyclic.
This is immediate from the lemma and from 2.4.4 applied to g.

Since (f g)S = g S f S we obtain:

g f
T 00 /T / T 0 morphisms of topologies, F 0 ∈ ST 0 :
E2p,q = Rp g S (Rq f S (F 0 )) =⇒ R∗ (f g)S (F 0 )
In the special case T 00 is the discrete category {X} (there is only one topol-
ogy), X ∈ T , and g is the inclusion map we have Rp q S F = H p (T, X; F ) (F ∈ ST )
and Rp (f g)S F 0 = H p (T 0 , f (X); F 0 ) (F 0 ∈ ST 0 ). Hence

f
T / T 0 a morphism of topologies, X ∈ T , Y = f (X), and F 0 ∈ ST 0 .
E2p,q = H ( T, X; Rq f S (F 0 )) =⇒ H ∗ (T 0 , Y ; F 0 )
29
Note: in case f is obtained from a map π : X 0 → X of topological spaces as

in 2.4.1 (so that Y = X 0 ) this is the usual Leray spectral sequence. However, one
needs much stronger conditions on f than those of 2.4.2 to get a spectral sequence
relating H q (T, ) and H q (T 0 , ) (cf. 2.2.2) in general. For one thing, lim Γ ◦ f S
←−
need not be isomorphic to the functor lim Γ0 = H 0 (T 0 , ).
←−
Definition 2.4.3. Let T be a topology, Y ∈ T . Define a topology T /Y by
T /Y = T /Y and Cov T /Y = set of families of maps {Uα → U } over Y such
that {Uα → U } ∈ Cov T .
Let ² : T /Y → T be the functor ”ignore Y”. ² is a morphism of topologies

and it is trivial that ²S is exact. Applying 2.4.2 and taking into account the fact
that (Y, idY ) is final in T /Y we find (elucidating the functors 2.2.1)
Corollary 2.4.7. Let T be a topology, Y ∈ T , and ² : T /Y → T the canonical

φ
morphism. Let F ∈ ST and U −→ Y ∈ T /Y. Then
H q (T, U ; F ) ≈ H q (T /Y, (U, φ); ²S F )
In particular,
H q (T, Y ; F ) ≈ H q (T /Y, (Y, id); ²S F )
Theorem 2.4.8. Let f : T → T 0 be a morphism and suppose
i T and T 0 have final objects and finite fibered products.
ii f preserves final objects and finite fibered products (i.e., f is left exact).
Then fS is exact.
Proof : We first show fP is exact. It suffices to show that the categories IVf ,
V ∈ T (cf. 1.3.1) satisfy axioms (L1,2,3)◦ dual to (L1,2,3) of Section 1.2.
(L1)◦ : If
(U2 , φ2 ) (U3 , φ3 )
KKK s
KKK
KKK sssss
s
K% sy ss
(U1 , φ1 )
30
2 SHEAVES
is given in IVf , i.e., a diagram
U2 A U3
AA }
AA }}
AA }}}
AÃ }~ }
U1
in T such that
u V IIII
φ2 uuu IIφ3
uu II
uuu II
uz $
f (U2 ) φ1 f (U3 )
HH v
HH vv
HH vv
HH vv
H$ ² vz v
f (U1 )
commutes, take the induced map
V → f (U2 ×U1 U3 ) ≈ f (U2 ) ×f (U1 ) f (U3 )
/
(L2)◦ Given (U2 , φ2 ) / (U1 , φ1 ) , i.e., a diagram
α /
U2 / U1
β
in T such that the two triangles in the diagram

φ2
V DD / f (U2 )
DD φ
DD 1
DD
D! ² ²
f (U1 )
κ
commute, denote by K −→ U2 the kernel of the pair (α, β) of maps, i.e., K is the
object making the diagram below cartesian
K II
uu II
κ uuu IIκ
u II
uu II
zuu $
U2 H U2
HH v
HH vv
HH vvv
Γα HH$ v
zvv Γβ
U2 × U1
31
(where the product U2 × U1 is over the final object).

/ f (U2 ) /
By assumption ii, f (X) / f (U1 ) is again exact, so there exists
a canonical map ψ such that the diagram
ψ
V DD / f (X)
DD
DD
DD
D! ²
f (U2 )
commutes. Hence we get (K, ψ) → (U2 , φ2 ) which is as required.

(L3)◦ Given (U1 , φ1 ), (U2 , φ2 ) take (U1 × U2 , φ1 × φ2 ).
Now we claim that, whenever f is such that the categories IVf (V ∈ T ) satisfy
(L1,2,3)◦ , fS is exact: Let 0 → F 0 → F → F 00 → 0 be exact, the sequence
]0 ◦fP [0 → F 0 → F → F 00 → 0] is exact in ST 0 and so we have to show (fP C)] = 0.
Let V ∈ T 0 and
ξ¯ inf C(V ) = lim C(U )
P −→
(U,φ)∈IVf
and choose ξ ∈ C(U ), (U, φ) ∈ IVf representing ξ¯ (we are tactily using the discus-
sion of Section 1.2, notably 1.2.3). Since C ] , and hence C + , is the zero sheaf, we
Q
can find {Uα → U } ∈ Cov T such that ξ Ã 0 in C(Uα ). Consider the diagram
V o V
v α
= V ×f (U ) f (Uα )
φν vvv
v
φ vv φα
² vzv ²
f (U ) o f (Uα )
where φα is the composition.

In the functor IVf → IVfα induced the map V ← Vα we have (U, φ) Ã (U, φα )
given by U ← Uα . Hence since ξ Ã 0 in C(Uα ) it follows that ξ¯ Ã 0 in fP C(Vα ).
But {Vα → V } is obtained by base extension from a covering, hence is in Cov T’.
Q
Since ξ¯ Ã 0 in fP C(Vα ) it follows ξ¯ Ã 0 in C + (V ), whence C + = 0.
Q. E. D.
32
2 SHEAVES
2.5 Inductive limits for nötherian topologies

Let T be a topology and P, S the categories of presheaves and sheaves on
T. If I is a category we write P I = HomCat (I, P), S I = HomCat (I, S). For
P ∈ P I , lim P is representable. In fact we need only take [lim P ](U ) = lim[P (U )].
−→ −→ −→
lim : P I → P is exact if I satisfies (L1,2) (cf. Section 1.2).
−→
LetF ∈ S I and let p. lim F denote the limit in the category of presheaves.
−→
Clearly, (p. lim F )] = lim F represents the functor lim F . We claim if I satisfies
−→ −→ −→ `
(L1,2) then lim : S I → S is exact. In fact, writing I = Iν as direct sum
−→ ν
of its connected components, and recalling that AB5 holds in S, one reduces to
the case I satisfies (L1,2,3). Moreover, lim is certainly left exact since we have
−→
lim = ] ◦ p. lim ◦iI (iI : S I ,→ P I ) and all the functors on the right are left exact.
−→ −→
So what has to be shown is that if P ∈ P I satisfies P ] = 0 (i.e., Pi] = 0 for all
I
i ∈ I, whence Pi+ = 0) then P ]

→ = (lim
− P )] = 0. But if ξ ∈ −
P (U ) is represented
→
−→ Q
by ξ ∈ Pi (U ) (cf. 1.2.3!) then ξ → 0 in Pi (Uα ) for some {Uα → U } ∈ Cov T
Q
since Pi+ = 0 and therefore ξ → 0 in − P (Uα ).
→
Definition 2.5.1. Let T be a topology. T is nötherian iff. every {Uα → U } ∈
Cov T has a finite subcover, i.e., a finite subfamily of {Uα → U } is in Cov T.
Note that this is quite rare.
Proposition 2.5.1. Let T be a nötherian topology, and let T f be the follwoing

topology: T f = T , Cov T f = set of families {Uα → U } ∈ Cov T which are finite.
Denote by S, S f the categories of sheaves on T, T f respectively. Then S = S f .
Proof : Obviously S ⊂ S f . Let F ∈ S f . We have to show that if {Uα → U } ∈

∼/ Q /
Cov T then F (U ) ker( F (Uα ) / F (Uα ×U Uβ )) . Let {Uα0 → U } be a
∼/ Q /
finite subcover of {Uα → U }. Then F (U ) ker( F (Uν0 ) / ) . Since this
Q / Q
/ ker( F (U 0 ) /
map factors through the canonical map ker( F (Uα ) / ) ν / )
ÂÄ Q /
it follows F (U ) / ker( F (Uα ) / ) , i.e., F satisfies (+) (cf. 2.1.1). Hence
Q / Â Ä Q /
by 2.1.3 ker( F (Uα ) / ) / ker( F (U 0 ) / ) and the result follows.
ν
Q. E. D.
33
2.5 INDUCTIVE LIMITS FOR NÖTHERIAN TOPOLOGIES
Proposition 2.5.2. Suppose T nötherian and I a category satisfying (L1,2). Let

F ∈ S f . Then
i p. lim F is a sheaf.
−→
ii If each Fi (i ∈ I) is flask in T f then lim F is flask in T f , where T f is as in
−→
2.5.1.
Proof : By 2.5.1 we may assume T = T f . Write − F = p. lim F and let {Uα →

→ −→
U } ∈ Cov T be a (finite) family. Now it is trivial that lim commutes with direct
−→
sums. Therefore
Y M
F (Uα0 ×U · · · ×U Uαn ) =
−
→ F (Uα0 ×U · · · ×U Uαn )
−
→
(α0 ,··· ,αn ) (α0 ,··· ,αn )
= lim(F (Uα0 ×U · · · ×U Uαn ))

−→
I
Y
= lim( F (Uα0 ×U · · · ×U Uαn ))
−→
I α
Since lim is exact by 1.2.4 the proposition follows immediately.

−→
I
Q. E. D.
Corollary 2.5.3. Let T be a nötherian topology, and I a category satisfying

(L1,2). The functors H q (T, U ; ) commute with lim.
−→
I
Note that given F : I → S we get functors hq : I → Ab by hq (i) =

H q (T, U ; Fi ). Since by 2.5.1 we may assume T = T f and since H q (T, U ; ) may
be computed with flask resolutions (cf.2.4.3) we are done by 2.5.2 if we can find
a resolution 0 → F → F 1 → F 2 → · · · of the functor F such that Fiq is flask all
q > 0, i ∈ I. In fact there exist resolutions with Fiq injective. This can be seen
without calculation as follows: P I and S I have all the axioms that P, S do, so
S I has enough injectives and we take for F q an injective resolution of F. If J is
φ
another category and J −→ I we get functors
φP
J o
P / PI
φP
34
2 SHEAVES
adjoint, which are constructed as with preschemes (cf. Section 2.2) (so φP (P ) =
P ◦ φ). In particular if J = {i} is the discrete category, i ∈ I and φ : {i} → I is
the inclusion and if P is a presheaf = element of P {i} then
M
(φP P )j = P
HomI (i,j)
Similarly, we get
φS
J o
S / SI
φS
with φS (F ) = F ◦φ (F ∈ S I ), and with φS = ]I ◦φP ◦iJ . Therefore if s ∈ S = S {i}

we have
M M
φS (S)j = (presheaf S)] = S
HomI (i,j) HomI (i,j)
Since AB5 holds in S it follows that φS : S → S I is exact and therefore that φS

carries injectives into injectives. Since φS F = Fi in this case. we are done.
A special case of 2.5.2, 2.5.3 is
Corollary 2.5.4. If T is nötherian then

L
i ”presheaf Fν ” is a sheaf.
ν
ii H q (T, U ; ) commutes with ⊕.
35
2.5 INDUCTIVE LIMITS FOR NÖTHERIAN TOPOLOGIES
36
3 THE ÉTALE GROTHENDIECK TOPOLOGY FOR SCHEMES
3 The étale Grothendieck topology for

schemes
3.1 Definition
Let X be a prescheme and consider the following topologies:
(f) TX0 = Category of preschemes Y /X étale, separated and finitely presented.

Cov TX0 = Finite families {Uα → U } of maps which are surjective, i.e., such
that U is covered by the union of the images of the Uα s.
(0) TX0 = Preschemes Y /X étale, separated, finitely presented.

Cov TX0 = Arbitrary surjective families {Uα → U }.
(1) TX0 = Preschemes Y /X étale, separated.

(2) TX0 = Preschemes Y /X étale.

We refer to these as cases (f), (0), (1), (2) respectively, and we are primarily
interested in cases (f) and (1). The definition of étale used is that of [2] IX, so
in particular étale implies locally of finite presentation. Notice that the maps in
TX have in each case the properties required of the structure maps, and that TX
is closed under fibred products (in the category of preschemes).
β
We have inclusions TXf
α /
TX0 / T 1 γ / T 2 which are obviously mor-
X X
phisms of topologies in the sense of 2.4.2. Denoting by S z the category of abelian
γS βS αS
sheaves on TXz (z = f, 0, 1, 2), we get functors S 2 / S1 / S0 / S f where
αS is an inclusion.
37
3.1 DEFINITION
Theorem 3.1.1.
i If X is quasi-compact then αS is the identity.
ii If X is quasi-separated then β S is an equivalence of categories.
iii γ S is an equivalence of categories.
Proof : i follows from 2.5.1 and the following
Lemma 3.1.2. If X is quasi-compact then TX0 is nötherian.
Proof of 3.1.2: Let {Yα → Y } ∈ Cov TX0 . We have to show that already a finite
number of images cover Y. Since Y /X is finitely presented, it is quasi- compact.
Hence Y is quasi-compact. Also Yα /Y is étale, separated finitely presented. So
we may assume Y = X. Now one sees easily that an étale morphism is open,
using the fact that this is true if X is nötherian (cf. [2] IV, remark following
Theorem 6.6). Hence the image of each Yα is open. Since X is quasi-compact, we
are done.
Lemma 3.1.3. Let i : T 0 ⊂ T be topologies (i a morphism) and suppose
i T 0 is a full subcategory of T .
ii If {Uα → U } ∈ Cov T , and Uα , U ∈ T 0 for all α then {Uα → U } ∈ Cov T 0 .
iii Given U ∈ T , ∃{Uα → U } ∈ Cov T with Uα ∈ T 0 , all α. Then iS : ST →

ST 0 is an equivalence of categories.
Applying the lemma, we are reduced to checking (1), (2), (3) for β, γ with X
quasi-separated in case β. (1), (2) are trivial. For (3), recall that if Y /X is locally
finitely presented then for every y ∈ Y there is an open V ⊂ Y containing y with
V /X locally finitely presented and separated. For, by definition ∃V ⊂ Y and
U ⊂ X affine with V /U finitely presented (and certainly separated). Since U/X
is obviously locally finitely presented and separated, so is V /X. Hence if Y ∈ TX2
we can find a Zariski open covering in TX1 . Finally, if X is quasi-separated, and
38
V, U are as above then U ⊂ X is quasi-compact, hence also finitely presented

and so V /X is finitely presented and separated. Hence the Zariski open covering
is in TX0 in this case.
Q. E. D.
Proof of 3.1.3: Since the functors iS : ST → ST 0 and iS : ST 0 → ST are adjoint,

there are canonical morphisms of functors φ : iS iS → idST and ψ : idST 0 → iS iS .
We have to show these are isomorphisms. To show ψ is an isomorphism we need
∼
to show for F ∈ ST 0 , V ∈ T 0 that F (V ) −→ iS F (V ) = iS iS F (V ). Now clearly
∼
F (V ) −→ iP F (V ) since (V, idV ) is an initial object in the category IVi (cf. 1.3.1).
Applying the assumptions of the lemma, we find that the coverings from T’ of
V are initial (final in the dual category) in the category of coverings of V in T,
∼
and since F is a sheaf on T’ we find iP F (V ) −→ (iP F (V ))] = iS F (V ). So we are
done.
∼
To show φ is an isomorphism we want iS iS G(U ) −→ G(U ) for U ∈ T ,
G ∈ ST .
Case (i): U ∈ T 0 . Then U, idU is initial in IUi , so
∼
iP iP G(U ) = lim iS G(U ) = lim G(V ) −→ G(U )
−→ i −→
(V,φ)∈IU
∼
Again (iP iP G(U ))] = iS iS G(U ) −→ G(U ) because we only have to look at cov-
erings from T’ and we have isomorphism there.
Case (ii): U ∈ T arbitrary. Applying (3), we may choose {Uα → U } ∈ Cov T
with Uα ∈ T 0 . Choose also for each pair α, β of indices {Vαβ
ν
→ Uα ×U Uβ } ∈ Cov T
ν
with Vαβ ∈ T 0 . Then for any S ∈ ST we have
Y Y γ
S(Uα ×U Uβ ) ,→ S(Vαβ )
α,β α,β,ν
Hence in the diagram below the rows are exact. Since by case (1) the two right
vertical arrows are isomorphisms (the arrows induced by φ), so is the left one.
39
3.1 DEFINITION
This completes the proof of the lemma.

Q /Q
0 / i iS G(U )
S
/ iS iS G(Uα ) / iS iS G(Vαβ
ν
)
∼
= ∼
=
² ² ²
/ G(U ) /
Q /Q ν
G(Vαβ )
0 G(Uα ) /
Q. E. D.
Proposition 3.1.4. The coverings in the cases (f), (0), (1), (2) are universal
effective epimorphisms in the category of preschemes.
`
We omit the proof. Once reduces easily to the case Uα → U faithfully flat
and quasi-compact (cf. [2] VIII Corollary 5.3).
Let π : X → Y be a morphism of preschemes. π induces a morphism of
topologies f (π) = f : TYz → TXz (z = f, 0, 1, 2) by f (U ) = U ×Y X for U ∈ TYz .
Notation 3.1.1. Assuming Z known and fixed we write
SX = STXz = category of abelian sheaves on TXz
PX = PTXz = category of abelian presheaves on TXz
π∗ = f S
π ∗ = fS
Since f preserves final objects and fibred products we have by 2.4.8
Corollary 3.1.5. π ∗ = fS is an exact functor.
Suppose P = spec k̄ where k̄ is a separably algebraically closed field. Then

TPz is equivalent with the canonical topology (cf. 1.1.2) on the category of sets
(resp. finite sets if z = 0, resp. finite sets, finite covering families if z = f). In
any case, SP is equivalent with Ab, the equivalence being given by G Ã G(P ),
G ∈ SP .
Definition 3.1.1. Let ² : P → X be a geometric point of X, i.e., P = spec k̄

with k̄ separably algebraically closed, and let F ∈ SX . The stalk of F at P is the
abelian group FP = ²∗ F (P ).
40
The functor F Ã FP is exact by 3.1.5 and the above remarks. One sees
easily, moreover, that if f : TX → TP is the morphism associated to ² then
²∗ F (P ) = fS F (P ) = fP F (P ), i.e., FP = lim F (V )
−→ f
(V,φ)∈IP
Since (X, idP ) ∈ IPf we get a canonical homomorphism F (X) → FP
Proposition 3.1.6. Suppose X quasi-compact in case (f), and let F ∈ SX , ξ ∈

F (X). Then ξ = 0 iff. the image of ξ in FP is zero for every geometric point P.
Proof : ⇒ is trivial. So suppose ξ Ã 0 in every FP , and let x ∈ X. Choose

² : P → X lying over x. Since IPf satisfies (L1, 2, 3)◦ (cf. proof of 2.4.8) we may
argue as follows: There is some U/X ∈ TX and a map φ : P → U commuting
with ² so that ξ Ã 0 in F(U). Then the image of U on X contains x. Since this
can be done for each x, we can find {Uα → U } ∈ Cov TX (finite in case (f)) with
Q
ξ Ã 0 in F (Uα ). Therefore ξ = 0.
Q. E. D.
Proposition 3.1.7. Let F be a presheaf on X, P → X as above a geometric

point. Then fP F (P ) ≈ fP F ] (P ) ≈ FP] , where the isomorphism is induced by the
canonical map F → F ] .
The proof is essencially obvious. For instance, to show the map surjective,
let ξ ∈ fP F ] (P ) be represented by ξ ∈ F ] (V ) for some φ : P → V commuting
Q
with ². Then there is a {Vα → V } ∈ Cov T , and ξ 0 ∈ F + (Vα ) representing ξ.
`
Since Vα → V is surjective we can lift φ to some map φα : P → Vα . Let ξα0 be
Q
the αth component of ξ 0 , and choose {Wβ → Vα } ∈ Cov T and ξ 00 ∈ F (Wβ )
representing ξα . Lift φα to some φβ : P → Wβ and let ξβ00 be the βth component of
ξ 00 . Obviously the image ξ 00 of ξβ00 in fP F (P ) (obtained with respect to (Wβ , φβ ) ∈
IPf ) is mapped onto ξ.
3.2 Relations with a closed subscheme

Throughout this section we assume we are in case (1). Let i : X → Y be a
closed subscheme and let F ∈ SY .
41
3.2 RELATIONS WITH A CLOSED SUBSCHEME
Definition 3.2.1. F is zero outside X iff. F (U ) = 0 whenever U ×Y X = ∅.
Clearly i∗ carries SX into the full subcategory S¯Y of SY consisting of sheaves

zero outside X.
Theorem 3.2.1. The functor SX → S¯Y induced by i∗ is an equivalence of cate-

gories.
Proof : Let ² : TX0 ⊂ TX be the full subtopology (i.e., all maps and coverings)
whose objects are preschemes V̄ /X where V̄ = V ×Y X for some V ∈ TY . We
claim TX0 satisfies the conditions of Lemma 3.1.3 and hence that we may replace
TX by TX0 . Condition (iii) is the only nontrivial one. Let Z/X ∈ TX , z ∈ Z. By
definition of étale, we can find z ∈ V̄ ⊂ Z, Ū ⊂ X affine with V̄ → Ū and affine
∼
V̄0 /Ū0 of finite type over specZ and maps Ū → Ū0 , V̄ −→ V̄0 ×U Ū0 . We may
assume Ū0 = U × X for some affine U ⊂ Y and then we can choose U → U0 and
Ū0 → U0 closed, with U0 of finite type over specZ. Let z0 ∈ V̄0 be the image of z.
We are reduced to the nötherian case, and it follows easily from [2] I. Theorem
7.6 and Proposition 4.5.
Replace TX by TX0 and let f : TY → TX0 be the morphism f (U ) = U ×Y X.
We want to show f S induces an equivalence f 0 : SX → S¯Y . The adjoint functor
f0 : S¯Y → SX is obviously obtained from fS by restricting to S¯Y , so we have
morphisms φ : f0 f 0 → idSX , ψ : idS¯Y → f 0 f0 which we want to show are
isomorphisms. Actually, it will appear that fP |S¯Y is the adjoint since fP F is a
sheaf for F ∈ S¯Y .
Let V̄ ∈ TX0 , V̄ = V ×Y X with V ∈ TY and consider the category
IV̄f = {(U, φ)|U ∈ TY , φ : U ×Y X ← V̄ }
We claim that the (U, φ)s in IV̄f with φ an isomorphism form an initial sub-
category. In fact, if (U, φ) is arbitrary then (U, φ) ← (V ×Y U, Γφ ) (identify-
ing V ×Y U ×Y X with V̄ ×X (U ×Y X)). Write Ū = U ×Y X. The graph
Γφ : V̄ → V̄ ×X Ū is open and closed ([2] IX. Corollary 1.6 and above), hence we
` ∼
may write V̄ ×X Ū = Γ Λ where V̄ −→. In the map V̄ ×X Ū → V ×Y U (which
42
is a closed immersion), Λ is mapped onto a closed subset. Set W = U ×Y V − Λ.

Then W ∈ TY and we get a map (U, φ) ← (W, ψ) with ψ an isomorphism.
Next, we claim that for F ∈ S¯T , V ∈ TX , F (V ) → fP F (V̄ ) is an isomor-
phism, where V̄ = V ×Y X and the map is given by (V, idV̄ ) ∈ IVf . This is trivial
if V ×Y X = ∅. Since (L1,2,3)◦ hold in IV̄f (cf. proof of 2.4.8) and by the above
discussion, we may replace IV̄f by the category of pairs (U, φ) in which φ is an
∼
isomorphism, and then it suffices to show F (U ) −→ F (U 0 ) for (U, φ) ← (U 0 , φ0 )
(and φ, φ0 isomorphisms). View Ū = U ×Y X as a closed subscheme of U. Since
φ is an isomorphism, U’ covers Ū and therefore covers some open set containing
it. Hence we may separate the two cases
1. U 0 ⊂ U is an open subset.
2. U 0 → U surjective, i.e., {U 0 → U } ∈ Cov TY .

`
Case 1. Set U 00 = U − Ū , so U 0 U 00 → U is surjective. Then U 00 ×Y X = ∅,
hence F (U 00 ) = 0 and we are reduced to case 2.
Case 2. Consider
p1
o
Uo U0 o p2
U 0 ×U U 0
and let ∆ ⊂ U 0 ×U U 0 be the image of the diagonal. ∆ is open and closed in

`
U 0 ×U U 0 , so we can write U 0 ×U U 0 = ∆ Λ. Since φ and φ0 are isomorphisms,
∼
U 0 ×Y X −→ U 0 ×Y X we find Λ ×Y X = ∅. Hence F (Λ) = 0. But p1 = p2 on
/
∆. Therefore F (U 0 ) / F (U 0 ×U U 0 ) is the zero map and since F is a sheaf,
∼
F (U ) −→ F (U 0 ) as required.
Now it is clear that fP F is a sheaf. For , if {V̄α → V̄ } ∈ Cov TX0 , V̄α =

Vα ×Y X, V̄ = V ×Y X, we may replace V̄α by isomorphic elements of TX0 , say V̄α0
so that V̄α0 → V̄ is induced by a map Vα0 → V . Since {Vα0 } cover V̄ , they cover
some open neighborhood of V̄ . Hence we may assume {V̄α → V̄ } is induced from
a covering {Vα → V } ∈ Cov TY . Since F is a sheaf we are done.
The fact that φ : f0 f 0 → idSX and ψ : idS̄Y → f 0 f0 are isomorphisms is now
immediate. Q. E. D.
By transitivity of equivalence we have:
43
Corollary 3.2.2. The functor of Theorem 3.2.1 is an equivalence of categories

in case (2), and in case (0) if Y is quasi-separated, and in case (f) if Y is quasi-
compact and quasi-separated.
f
Let for the moment A → B be any left exact functor, A, B abelian categories.
Construct a new category C as follows: ObC = set of triples (B, A, φ) where
ξ
B ∈ B, A ∈ A, φ ∈ Hom(A, f (B)). A map (B, A, φ) → (B 0 , A0 , φ0 ) is a pair
ξ ξ
B
B −→ B 0 and A −→
A
A0 of maps such that
φ
A / f (B)
ξA f (ξB )
² φ0
²
A0 / f (B 0 )
commutes. Using the fact that f is left exact one verifies easily that C is again an
abelian category. C inherits most properties enjoyed by A, B.
Therefore are various functors relating A, B, C, notable the six
o j∗∗
!
o i
i∗ j
A i∗ / C j! / B
o o
defined as follows:
i∗ : A o o/ o/ o/ /o (B, A, φ) j! : (B, 0, 0) o o/ o/ o/ o/ /o B
i∗ : A /o /o /o /o / (0, A, 0) ; j ∗ : (B, A, φ) /o /o /o o/ / B
i! : kerφ o o/ o/ /o (B, A, φ) j∗ : (B, f (B), id) o o/ o/ /o B
We insert for reference a list of obvious properties:
i A given functor is left adjoint to the one below it.
ii i∗ , i∗ , j ∗ , j! are exact, j∗ , i! are left exact.
iii i∗ j∗ = f , i∗ j! = i! j! = i! j∗ j ∗ i∗ = 0.
iv i∗ , j∗ are fully faithful and for C ∈ C, j ∗ C = 0 iff C ≈ i∗ A, some A ∈ A.
44
Proposition 3.2.3. Let

i∗ / o j∗∗
A o i∗ C̄ j / B
be abelian categories and functors. Suppose the functors satisfy the following
conditions:
i i∗ (resp. j ∗ ) is left adjoint to i∗ (resp. j∗ ).
ii i∗ , j ∗ are exact.
iii i∗ , j∗ are fully faithful.
iv For C ∈ C̄, j ∗ C = 0 ⇐⇒ C = i∗ A, some A ∈ A.
Set f = i∗ j∗ , and denote by C the category of triples (B, A, φ) as above. Then C̄

is equivalent with C.
To be sure, the equivalence ’preserves’ the given functors. Since j∗ has a left
adjoint, it is left exact and so f is left exact.
Proof : Note first that for C ∈ C̄, C = 0 iff. i∗ C = 0 and j ∗ C = 0. For, say
j ∗ C = 0, when C ≈ i∗ A some A ∈ A. Since i∗ is fully faithful, the canonical map
i∗ i∗ A → A deduced from i is an isomorphism. So if also i∗ C ≈ i∗ i∗ A = 0 then
A = 0, hence C = 0. Since i∗ , j ∗ are exact we conclude that a map ² : C → C 0 in
C̄ is an isomorphism iff i∗ (²), j ∗ (²) are isomorphisms.
Let C ∈ C̄ be arbitrary. The adjointness properties i induce a commutative
diagram
i∗ i ∗ C o C
² ²
i ∗ i∗ j ∗ j ∗ C o j∗ j ∗ C
²
and hence a map C −→ (i∗ i∗ C) ×i∗ i∗ j∗ j ∗ C (j∗ j ∗ C). We claim ² is an isomorphism,
and we need to show i∗ (²), j ∗ (²) are isomorphisms. Both functors i∗ , j ∗ are exact
by assumption and therefore commute with fibred products. Since i∗ is fully
faithful, i∗ C → i∗ i∗ i∗ C is an isomorphism for all C ∈ C, hence the horizontal
arrows of the diagram become isomorphisms upon applying i∗ , so the square
45
certainly becomes cartesian. Applying j ∗ to the diagram we get zero on the left
∼
side by iv, and j ∗ C −→ j ∗ j∗ j ∗ C because j∗ is fully faithful, so again the square
becomes cartesian and we are done.
This argument applies of course also to the category C of triples, so we have
∼
(B, A, φ) −→ (0, A, 0) ×(0,f (B),0) (B, f (B), id).
Therefore define a functor F : C̄ → C by C Ã (j ∗ C, i∗ C, φ) where φ : i∗ C →
i∗ j∗ j ∗ C = f (j ∗ C) is induced by the map C → j∗ j ∗ C, and define G : C → C̄ by
(B, A, φ) Ã i∗ A ×i∗ j ∗ j∗ B j∗ B. It is immediate that one obtains an equivalence of
categories.
Q. E. D.
Corollary 3.2.4. Let i : X → Y be a closed subscheme and set U = Y − X,

j : U ,→ Y the map. The category SY of sheaves on Y is equivalent with the
category of triples (B, A, φ) where B ∈ SU , A ∈ SX and φ : A → i∗ j∗ B.
For if we set A = SX , B = SU , C̄ = SY in the above proposition then

condition i is satisfied by definition. For ii see 3.1.5 and for iv see 3.2.1. Condition
∼
iii for i∗ follows also from 3.2.1, and for j∗ we need only show B −→ j ∗ j∗ B,
B ∈ SU . This is clear (see for example 1.3.2).
In this context the functor i! associates with a sheaf on Y the subsheaf of
sections with support on X, and j! is the extension of a sheaf on U by zero. This
interpretation greatly facilitates the calculation, of the usual exact cohomology
sequences for closed subsets, for instance we have immediately.
For C ∈ SY , we have the exact sequence
0 → j ! j ∗ C → C → i ∗ i∗ C → 0 (3.2.1)
which, read in the category C is
0 → (B, 0, 0) → (B, A, φ) → (0, A, 0) → 0
We also have a left exact sequence
0 → i∗ i! C → C → j∗ j ∗ C, C∈C (3.2.2)
46
If C ∈ C is an injective then the last arrow is in fact surjective (say we work in

C, assuming A, B (hence C) have enough injectives). For, if C is injective, so are
i∗ i! C and j∗ j ∗ C. Hence C ≈ i∗ i! C × C 0 with C 0 also injective and j∗ j ∗ C ≈ j∗ j ∗ C 0 .
Therefore j∗ j ∗ C ≈ C 0 × C 00 . But since j∗ is fully faithful, j ∗ C 0 → j ∗ j∗ j ∗ C ≈
j ∗ C 0 × j ∗ C 00 is an isomorphism and so j ∗ C 00 = 0. Since also
∼
C 0 × C 00 ≈ j∗ j ∗ C −→ j∗ j ∗ j∗ j ∗ C ≈ j∗ j ∗ C 0 × j∗ j ∗ C 00
it follows that C 00 = 0 and we are done.

Therefore we obtain an exact cohomology sequence
· · · → Rq (i∗ i! )C → Rq idC → Rq (j∗ j ∗ )C → · · ·
We have Rq id = 0, Rq (i∗ i! ) ≈ i∗ Rq i! , Rq (j∗ j ∗ ) ≈ (Rq j∗ )j ∗ , and so we find
i∗ R1 i! C ≈ coker(C → j∗ j ∗ C)orR1 i! (B, A, φ) ≈ cokerφ (3.2.3)
i∗ Rq i! C ≈ (Rq−1 j∗ )j ∗ C, q>1 (3.2.4)
or, since i∗ i∗ ≈ idA and i∗ Rq j∗ ≈ Rq i∗ j∗ = Rq f
Rq i! (B, A, φ) ≈ Rq−1 f (B), q>1 (3.2.5)
Note that Rq j∗ (q > 0) has always the form (0, x, 0), i.e., in case of the closed
j
subscheme Y − X = U → Y , the sheaves Rq j∗ B are zero outside of X (q > 0).
Let F : C → D be a left exact functor (D another abelian category). Since
for C ∈ C injective the sequence 3.2.2 is exact and consists of injectives, we get an
exact sequence 0 → F i∗ i! (C) → F (C) → F j∗ j ∗ (C) → 0. Therefore for arbitrary
C ∈ C we get the exact sequence
· · · → Rq (F i∗ i! )C → Rq F C → Rq (F j∗ j ∗ )C → · · · (3.2.6)
i
Notation 3.2.1. For X → Y a closed subscheme, U = X − Y ,C ∈ SY , we write
0
HX (Y, C) = H 0 (X, i! C) = H 0 (Y, i∗ i! C)
q
HX (Y, ) = Rq (H 0 (X, i! ))
47
3.3 PASSAGE TO THE LIMIT
0 q
HX (Y, C) is the group of sections of C with support on X, and HX (Y, C) is
q
the relative cohomology. Of course, there is a spectral sequence relating HX (Y, )
with Rq i! . Taking into account the fact that j∗ is exact and carries injectives to
injectives. 3.2.6 reads (relative cohomology sequence)
q
· · · → HX (Y, C) → H q (Y, C) → H q (U, j ∗ C) → · · · (3.2.7)
This should be contrasted with the exact sequence arising from 3.2.1, viz.
· · · → H q (Y, j! j ∗ C) → H q (Y, C) → H q (X, i∗ C) → · · · (3.2.8)
Since j! does not in general carry injectives into injectives we can not write
H q (Y, j! j∗ ) = Rq (H 0 (Y, j! j ∗ )).
For the sake of completeness, we note that a left exact functor F : C → D
induces functors Fa : A → D and Fb : B → D, left exact, by Fa = F i∗ , Fb = F j∗ ,
and a morphism Ff : Fb → Fa f by j∗ B → i∗ i∗ j∗ B = i∗ f (B). Since F is left
exact, it commutes with products and therefore we have for C = (B, A, φ) ∈ C a
cartesian diagram
F (C) M F (C) L
q MMM r LLL
qqqqq MMM
rrrrr LLL
q MMM r LLL
qx qq M& ry rr %
F i∗ (A)M F j∗ (B) = Fa (A)L Fb (B)
MMM q LLL ss
MMM qqqqq LLL sss
MMM qq L ss
& x qq
q Fa (φ) LL% sy ss Ff (B)
F i∗ i∗ j∗ (B) Fa (f (B))
Ff
Hence F can be recovered if Fa −→ Fb f are known. Thus for instance, if we are
i
in the case of a closed subscheme X → Y , U = Y − X, the section functors ΓV
(cf. 2.2.1) for V ∈ TY are determined by the functors ΓV ×Y X on SX and ΓV ×Y U
on SU and by the canonical morphism
ΓV ×Y U → ΓV ×Y X i∗ j∗
3.3 Passage to the limit

Let J be a category. Recall the following definition: A pseudofunctor C :
J → Cat is
48
a a map ObJ → ObCat; j Ã Cj .

f fc
b a map F lJ → F lCat; i → j Ã Ci −→ Cj .
g f ∼
c For i → j → k an isomorphism of functors cf,g : fc gc −→ (f g)c
h g f
such that cf,id = cid,f = id and if i → j → k → l are three maps then
cf,gh ◦ (f ∗ cg,h ) = cf g,h ◦ (cf,g ∗ h)
(notation as in Godement).
Let D : J → Cat be another pseudofunctor. A morphism F : C → D is
a for j ∈ J a functor Fj : Cj → Dj .
∼
b for i → j an isomorphism of functors Ff : fd ◦ Fi −→ Fj ◦ fc such that for
g f
i → j → k in J
Ff g = cf,g Ff Fg d−1
f,g
The pseudofunctors and morphisms J → Cat form a category. Let X ∈ Cat,

and define the constant pseudofunctor constX = cX : J → Cat by (cX )i = X,
fcX = idX , (cX )f,g = id. It is clear that a functor X → Y induces canonically a
morphism cX → cY .
Definition 3.3.1. Let C : J → Cat be a pseudofunctor. We define a functor

lim C : Cat → Set by
−→
[lim C](X) = Hompsfnct (C, cX )

−→
with the obvious maps.
We assume J satisfies (L1,2) of Section 1.2 for simplicity (For a more general
discussion the reader may consult [2] VI). The point is that lim C is representable,
−→ L
i.e., there exists a category lim C = −
C
→ , and a morphism of pseudofunctors C →
−→
const C such that by composition with L
→
−
∼
HomCat (−
C , X) −→ Hompsfnct (C, cX ) = [lim C](X)
→ −→
49
`
−
→C may be constructed as follows: Set Ob−→ C = i Ob Ci . For U, V ∈ Ob− C
→
we have to define Hom C (U, V ). Say U ∈ Ob Ci , V ∈ Ob Cj and denote by i, j|J
→
−
the category of diagrams
i ==
==
==
==
Á
@k
£££
£
££
££
j
where a map
   
i == i >>
 ==   >> 
 ==   >> 
 ==   >> 
 Á   Â 
 @k  −−−→  k0 
 ££   ¢¢@ 
 ££   ¢¢ 
 ££   ¢¢ 
££ ¢¢
j j
is a map k → k 0 such that the resulting triangles commute. i, j|J satisfies

(L1,2,3) (cf. 1.2.6 which is not stated with sufficient generality). Define a functor
H(U, V ) = H : i, j|J → Sets by
i <<
<< g
<<
<<
À
/o o/ o/ / 0
¤ A k H Hom(gc (U ), gc (V ))
¤¤
¤¤¤g0
¤¤
j
and for
i <<
<< g
<<
<<
À f H(i,j|f )
/ l let Hom(gc (U ), gc0 (V )) / Hom(f gc (U ), f g 0 (V ))
Ak c
¤¤¤
¤
¤¤¤ g0
¤
j
50
be defined by the functor fc and the morphisms cf,g , cf,g0 . Specifically, for φ ∈
Hom(gc (U ), gc0 (V )), we have
H(i, j|f ) : φ Ã cf,g0 (V ) ◦ fc (φ) ◦ c−1

f,g (U )
This gives indeed a functor and so we set
Hom C (U, V ) = lim H = lim H(U, V )

→
− −→ −→
The composition of maps in −C is self explanatory, and one gets functors

→
Li : Ci → −
C by the inclusion on Ob Ci and the fact that
→
i ????
ÄÄÄÄ
i ∈ i, i|J
i
Together with suitable isomorphisms Lf these give the morphisms L : C →
const C . We omit a few pages of tedious verification.
→
−
Similarly, we define a pseudofunctor T : J → T op (cf. 2.4.2) to be a
pseudofunctor of J into the underlying categories of T op such that the morphisms
are in T op, i.e., T consists of a map
ObJ → ObT op (j Ã Tj )
a map
F lJ → F lT op (f Ã ft )
and isomorphisms tf,g : ft gt → f gt with the same axioms as above. A morphism

F : T → T 0 is a morphism of the underlying pseudofunctors T , T 0 : J → Cat
such that for i ∈ J , Fi : Ti → Ti0 is a morphism of topologies. For X ∈ T op, set
[lim T ](X) = Hompsfnct (T, cX )

−→
so that lim T is a functor T op → Sets.
−→
Proposition 3.3.1. Let J be a category satisfying (L1,2), T : J → T op a
pseudofunctor and suppose for all j ∈ J that the covering families of Tj are finite
families. Then lim T is representable.
−→
51
φα
Proof : Set −
T = lim[T ] where lim[T ] is as above. Let {Uα −→ U } be a family
→ −→ −→
of maps of −T
→ , say Uα ∈ T iα , U ∈ Ti . Put {Uα → U } in Cov −
T if it is a finite
→
k g g
family and if there exists j ∈ J , maps iα −→ j and i −→ j and elements φαj ∈
φαj
Hom(gα (Uα ), g(U )) representing φα such that {gα (Uα ) −→ g(U )} ∈ Cov Tj . It
is clear that with this definition axioms 1, 2 of 1.1.1 are satisfied. We have to
φα ψ
verify 3. So let {Uα → U } ∈ Cov − T , V −→ U ∈ −
→ T . Reflecting a moment on
→
the construction of −
T one sees that we may replace the objects and maps by an
→
isomorphic situation so that Uα , U, V ∈ Tj for some fixed j, so that φα , ψ are
represented by maps φα , ψ in Tj , and so that with these maps {Uα → U } ∈
α2 P
Cov Tj . Then since Tj is a topology Uα ×U V exists and {Uα ×U V −→ V}∈−
T
→
is in Cov −
T . It remains to show that ”Uα ×U V ” is the product in −
→ T , i.e., that
→
for W → U ∈ − →T , HomU (W, Uα ×U V ) ≈ HomU (W, Uα ) × HomU (W, V ). For this
purpose we may assume furthermore (fixing W and replacing by an isomorphic
situation) that W ∈ Tj and W → U is represented in Tj . Thus the desired fact
follows easily from the commutativity of lim with products (cf. 1.2.7) and the
−→
following
Lemma 3.3.2. Let W → U ← V ∈ Tj . Then

T ∼
Hom−
→
U (W, V )longlef tarrow lim Homft (U ) (ft (W ), ft (V ))
−→
f
j →k∈j|J
where the maps W → U ← V in − T are represented by the given ones in Tj and

→
the functor j|J → Sets is defined as before (although j|J 6= j, j|J ).
The proof is routine.

Hence −
T is a topology, and it is clear that the functors Li : Ti → −
→ T are
→
morphisms of topologies and that (−
T , L) represents lim T .
→ −→
Q. E. D.
g
Let T : J → T op be given and fix 0 ∈ J , F a presheaf on T0 . For 0 → j in
J we get a presheaf (gt )p F on Tj , and using the functor L0 : T0 → −
T we get a
→
presheaf −
F = (L0 )p F on −
→ →T . We have, for U ∈ −T,−
→ F (U ) = lim F (X) where
→ −→ L
(X,φ)∈IU 0
52
IUL0 is the category of pairs (X, φ), X ∈ T0 , φ ∈ Hom T (U, X) (cf. ??). Say U
→
−
was in Tj . Then for
0 <<
<< g
<<
<<
Á
@ j ∈ 0, i|J
£££
£
££ g0
££
i
we have
[(gt )p F ](gt0 (U )) = lim F (X)
−→
(X,φj )
where the limit is over (X, φj ) ∈ Igg0t(U ) (the category of pairs (X, φj ), X ∈ T0 ,
t
f
φj ∈ Hom(gt0 (U ), gt (X))). Suppose j → k in J . Define a functor
f¯ : Igg0t(U ) → Iffgg0t(U ) by (X, φj ) Ã (X, φk ) (3.3.1)

t t
where
φk = tf,g (X) ◦ ft (φj ) ◦ t−1
f,g 0 (U )
and by the identity map on morphisms. Once obtains in this way a functor
IU : 0, i|J → Cat carrying
0;
;;
;;g
;;
;À
/o /o /o / I g0t
A¥ j gt (U )
¥¥
¥
¥¥¥ g0
¥
i
∼
and it is easily seen that lim IU −→ IUL0 (where the lim is in the sense of honest
−→ −→
functors, not pseudofunctors).
The case in which we are interested is the following: Fix a topology T0 with
finite covering families, and a category J satisfying (L1,2,3). Let V : J 0 → T
be a functor (i Ã Vi ). We suppose that for i, j ∈ J products of the form
Vi ×Vj U , Vi × U exist for U ∈ T0 , and hence obtain a pseudofunctor T : J → T op
f
by i Ã T /Vi (cf. 2.4.3), and for i → j letting ft : Ti → Tj be the functor
53
U/Vi Ã U ×Vi Vj /Vj . The ft ’s are morphisms of topologies and the isomorphisms
tf,g are given by formula (3.3.9.1) of [1] I for fibred products. To avoid overloading
notation we denote also by J the category obtained from J by adjoining an
gt
initial element 0, and we extend T by sending 0 Ã T0 with T0 −→ Ti defined by
U Ã U × Vi . The topologies Tj all have finite covering families and so lim T = −
T
→
−→
exists.
Let as above L0 : T0 → − T be the functor and for a presheaf on T0 write
→g
F = (L0 )p F , Fi = (gt )p F for 0 → i. Notice that 0, i|J is canonically isomorphic
−
→
to i|J .
Theorem 3.3.3. Make the above assumptions and notations:
i Let U ∈ −
T , say U is in Ti . Then for a presheaf F on T0 ,
→
F (U ) ≈
−
→ lim Fi (gt0 (U )) ≈ lim F (U ×Vi Vj )
−→ −→
g 0 i→j∈i|J
i−→j∈i|J
ii If F is a sheaf then −
F is a sheaf, and if F is flask so is −
→ F.
→
iii The functors (L0 )p and (L0 )s ≈ (L0 )P |S0 are exact (S0 =sheaves on T0 ).
iv If F is a sheaf and U ∈ T0 then
H p (− F ) ≈ lim H p (Tj , U × Vj , Fj ) ≈ lim H p (T0 , U × Vj ; F )

T , U; −
→ → −→ −→
j∈J j∈J
Proof : We have first to explain the categories over which the limits are taken. To
f
begin with, note that for i → j the functors (ft )p has a very special form due to
the fact that ft has as left adjoint the functor ² : U/Vj Ã U/Vi (by composition
with Uj → Ui ). In fact, by 1.3.2 we have canonically (ft )p ≈ ²p . Thus if the
g0
functor i −→ j Ã F (U ×Vi Vj ) is defined in the obvious way we obtain also a
g0
functor i −→ j Ã Fi (gt0 (U )) and so an isomorphism of the limits of i. This second
functor is the same as the one induced by 3.3.1. The second isomorphism of iv
is clear from 2.4.7 if the functor are defined using those of i by universality of
H p (nb. 0|J = J ), and the first isomorphism of iv will be trivial if i, ii, iii are
known.
54
Revert to the notation of 3.3.1. The categories Igg0t(U ) satisfy (L1,2,3)◦ , indeed,
t
they have initial objects (cf. 1.3.2), and it follows readily that IUL0 also satisfies
(L1,2,3)◦ . This proves iii (cf. proof of 2.4.8). The functors (Igg0t(U ) )◦ → Ab,
t
((X, φj ) Ã F (X)), are clearly obtained by composition with the functor IUL0 →
Ab, ((X, φ) Ã F (X)). So to complete the proof of i one has only to check
Lemma 3.3.4. Let C : B → Cat be a functor, suppose B and the Ci satisfy

(L1,2,3) and set −
C = lim C. Let F : −
→ C → D be a functor and define Fi : Ci → D
→
−→
by composition. Then
∼
lim(lim Fi ) −→ lim F
−→ −→ −→
B Ci C
−
→
To prove ii, let F be a sheaf, and {Uα → U } ∈ Cov −→T . There are finitely
many α, and so we may replace the objects and maps by an isomorphic situation
so that the Uα , U are in T0 and the maps are represented in T0 with {Uα → U } ∈
Cov T0 . Then we have for all Vj , {Uα × Vj → Vj } ∈ Cov T0 . Hence since F is a
sheaf
Q /Q
/
F (U × Vj ) α F (Uα × Vj ) / F (Uα × Uβ × Vj )
= / ⊕F (Uα × Vj ) /
F (U × Vj ) / ⊕F (Uα × Uβ × Vj )
is exact. Since lim commutes with ⊕ and is exact for (L1,2,3) it follows from i
−→
that − F
→ is a sheaf. The flaskness is obtained in the same way. (One can also show
Q
F
−
→ is a sheaf of sets if F is by using 1.2.7 instead of = ⊕).
Q. E. D.
Now let X : J ◦ → Preschemes be a functor (j Ã Xj ). Assume still

J satisfies (L1,2,3). Denote by SXj = Sj the category of preschemes over Xj
separated and of finite presentation. Given i → j, so that Xi ← Xj , one gets
a functor Si → Sj by base extension, and hence a pseudofunctor S : J → Cat
(j Ã Sj ). If for all i → j in J the maps Xi ← Xj are affine, one can show
that lim X is representable in the category of preschemes, and affine over the
−→
Xj ’s. Let ←
X
− = lim X and call S X the category of preschemes over ←
X
− separated,
←− ←−
55
of finite presentation. Base extension to ←

X
− yields a morphism S → constS←
X
of
−
pseudofunctors, hence a functor lim S → S X .
−→ ←−
Theorem 3.3.5. Let X : J ◦ → Preschemes be a functor, J satisfying (L1,2,3).

Suppose that for j ∈ J , Xj is quasi-compact and quasi-separated, and that for
i → j in J the maps Xi ← Xj are affine. Let ← X
− = lim X. Then with the above
←−
notation the functor lim S → S X is an equivalence of categories.
−→ ←−
The proof will be in [1] IV (if J is inductive, which is enough for the appli-
cations).
Let Tj be the case (f) topology on Xj , T X the case (f) topology on ←
X
−. One
←−
obtains in the same way a pseudofunctor T : J → T op, (j Ã Tj ) and a morphism
of topologies lim T → T X .
−→ ←−
Theorem 3.3.6. Make the same assumptions as in 3.3.5. Then lim T → T X is

−→ ←−
an equivalence of topologies.
We mean by an equivalence of topologies F : T1 → T2 a morphism which

is an equivalence of categories T1 → T2 and such that a family {Uα → U } of T1
is in Cov T1 if and only if {F (Uα ) → F (U )} is in Cov T2 . It is clear that an
equivalence induces equivalences on the categories of presheaves and sheaves.
Proof of Theorem 3.3.6: Ti is a full subcategory of Si and T X is a full subtopology
←
−
of S X . Hence lim T is a full subcategory of lim S. Since obviously the morphism
←− −→ −→
lim T → T X is induced by the functor lim S → S X which is an equivalence, we
−→ ←− −→ ←
−
know anyhow that lim T → T X is fully faithful. We have to check
−→ ←
−
a Given U ∈ T X , ∃j ∈ J , Uj ∈ Tj and an isom. Uj ×Xj ←

X
− ≈ U.
←−
b If {U → V } ∈ Cov T X then ∃j ∈ J , {Uj → Vj } ∈ Cov Tj and isomorphisms

←
−
Uj ×X ←
X
− ≈ U , Vj ×Xj ←
X
− ≈ V which commute with the induced maps. (Since
the covering families are finite we have replaced {Uα → V } by the single
`
map Uα = U → V .)
To show a, one reduces readily to the case Xi = specAi , ← X− = spec−→A, U =

specB affine. Then choose Bi /Ai finitely presented, some i, so that Bi ⊗Ai −
A ≈ B,
→
56
and set Bj = Bi ⊗Ai Aj for i → j. A theorem of Grothendieck guarantees that

Bj /Aj is flat for some j. We need Bj /Aj étale, i.e., we have to show Ω1Bj /Aj =
Ω1j = 0 for some j (and we know Ω1B/ A = 0). But since Bj /Aj is of finite type,
→
−
so is Ω1j (as Bj -module). We have Ω1B/ A ≈ Ω1i ⊗Ai − A ≈ lim Ω1i ⊗Ai Aj . If {ων }
→
→
− −→
are a set of generators for Ω1i , so that {ω ⊗ 1j } generate Ω1j it is clear that these
elements must be zero for some j, as desired.
To show b, let U → V be surjective in T X , and choose i, Ui → Vi in Ti with
←
−
(Ui → Vi ) ×Xi ←
X
− ≈ U → V using a. Since Ui → Vi is étale, finitely presented, the
image Ṽi → Vi is also finitely presented, so Ui → Ṽi ∈ Cov Ti . Let Ṽ = V ×Xi ←
X
−.
Clearly Ṽ ≈ V , hence replacing Vi by Ṽi we are done.
Corollary 3.3.7. Let X0 be a prescheme, X : J ◦ → TX0 (j Ã Xj ) a functor,
where the topology is any of the cases (f,0,1,2). Suppose the Xj are affine. Sup-
pose finally J satisfies (L1,2,3). Let ←
X
− = lim X and for a sheaf F on X0 , let Fi ,
←−
F be the sheaves induced on Xi resp. ←
−
→ X
−. Then
∼
lim H q (Xi , Fi ) −→ H q (X
−, −
← F)
→
−→
J
Clear from 3.1.1, 3.3.3 and 3.3.6. Note that if π : ←

X
− → X is the map,
F = π∗F .
−
→
3.4 Hensel rings

We include for convenience some basic facts on Hensel rings. Most of the
results and proofs are taken from paper of [5] and [11].
Let A be a local ring, m its maximal ideal. Consider the category E(A) of
rings B local over A, B obtained by localizing an A-algebra of finite presentation
with B/A étale and B/mB ≈ A/mA. This category is an inductive system (cf.
[2] I. Corollary 5.4). Set
Ã = lim B
−→
B∈E(A)
Ã is naturally a functor of A. Clearly Ã is local, Ã/A is étale, and Ã/mÃ ≈ A/mA.

Ã has the following property:
If B/Ã is local, étale,localized from an Ã-algebra of
(3.4.1)
finite presentation, and if B/m̃B ≈ Ã/m̃Ã then Ã ≈ B
57
3.4 HENSEL RINGS
Or, equivalently,
If f ∈ Ã[t] is monic, and if ā ∈ Ã/m̃Ã[t] (the
(3.4.2)
image of f), then f has a simple root a inducing ā
To see that 3.4.1 =⇒ 3.4.2, set B = (Ã[t]/f )q where q is the maximal ideal
induced by the root ā of f¯. We get B ≈ Ã. Hence the image of t in B yields the
desired root in Ã. For the converse, use [2] I, Theorem 7.6 to write B ≈ (Ã[t]/f )q
where f is monic and q is a maximal ideal over m̃ (the theorem extends to the
non-nötherian case, if finite presentation is substituted for finite type). Obviously
B/m̃B ≈ Ã/m̃Ã means f¯ has a simple root. Hence f has a simple root inducing
it and one finds Ã[t]/f ≈ Ã × C for some C, so B ≈ Ã. It is now clear that Ã
has property 3.4.2.
Theorem 3.4.1. If A is nötherian, so is Ã.
This elegant proof is due to [11]: Clearly A/mn A ≈ Ã/m̃n Ã, hence A and Ã
have the same completion Â. Since Ã is a limit of nötherian local rings (which
are separated for the m = adic topology), A is separated, so we have A ⊂ Ã ⊂
Â. One uses the following criterion for nötherian rings: every ascending chain
· · · ⊂ ai ⊂ · · · of finitely generated ideals becomes constant. Since Â is nötherian
the chain · · · ⊂ ai Â ⊂ · · · of ideals of Â becomes constant, hence the chain
· · · ⊂ ai Â ∩ Ã ⊂ · · · becomes constant. So it suffices to show aÂ ∩ Ã = a for
finitely generated ideals a, and it is obvious that aÂ ∩ Ã ⊃ a. Let {a1 , · · · , an }
generate a, and let b ∈ aÂ ∩ Ã. Choose B/A ∈ E(A) so that a1 , · · · , an , b ∈ B
and set a0 = (a1 , · · · , an )B. Since B is nötherian with completion Â, it is well
known that a0 Â ∩ B = a0 . But a0 Â ∩ B = aÂ ∩ B contains b. Hence b ∈ a0 and
so b ∈ a.
Q. E. D.
Theorem 3.4.2. Let A be a local ring, m its maximal ideal, Ā = A/mA. The
following are equivalent:
i If f ∈ A[t] is a monic polynomial such that its image f¯ ∈ Ā[t] has the form
f¯ = ḡ h̄ with ḡ, h̄ monic and (ḡ, h̄) = 1 there exist unique monic polynomials
g, h ∈ A[t] representing ḡ, h̄ respectively such that f = gh, (g, h) = 1.
58
ii Every finite A-algebra is a direct product of local rings.
iii A (= Ã) has property 3.4.1.
iv Definition: A is a Hensel ring.
Proof : i =⇒ ii: Let B/A, finite, be given, and decompose B̄ = B/B (which is
finite over the field Ā) into a direct product of local rings. We have to find a
corresponding decomposition of B into a direct product. So if ē is an idempotent
of B̄, we have to find an idempotent of B lying over ē. Let b ∈ B be any element
lying over ē. Clearly we may assume b generates B. Then B̄ ≈ Ā[t]/f¯ for some
monic polynomial f¯ hence by Nakayama B is a quotient of A[t]/f with f monic
representing f¯. But the decomposition of B̄ into a direct product corresponding
to the factorization of f¯ into relatively prime factors. Since i holds, f factors
correspondingly, so we are done.
ii =⇒ iii: If B/A is given so as to test 3.4.1, write B ≈ Cq where C/A is
finite. Applying ii, C ≈ B × D for some D. Hence B is finite (and étale). Since
A/mA ≈ B/mB, B = A by Nakayama.
iii =⇒ i: Suppose A satisfies 3.4.1. Write Z[|A|] → A → 0 where |A| denotes
the set of elements of A, and let p ⊂ Z[|A|] be the inverse image of m. Set
A0 = Z[|A|]p , so that A0 → A is surjective and local. Since A ≈ Ã, we get
Ã0 → A. Now it is clear that i is preserved in quotients, so we are reduced to
showing i for Ã0 . Let f ∈ Ã0 [t] be given, monic, with f¯ = ḡ h̄, (ḡ, h̄) = 1 in Ā0 [t].
f is represented in B 0 [t] for some B 0 /A0 , B 0 ∈ E(A0 ). Applying [2] I, Theorem
7.6, we may write B 0 ≈ Cq0 where C 0 = A0 [t]/q(t). Choose finitely many elements
a1 , · · · , an of |A| so that, if A0 = (Z[a1 , · · · , an ])p0 where p0 is the prime ideal
induced by p, the coefficients of q(t) are in A0 , q̄(t) has the simple root ā in Ā0
inducing q, and f¯(t) factors as in Ā0 [t]. Set B0 = (A0 [t]/q(t))q0 (q0 induced by
(t − ā)). Then B0 ⊂ Ã0 . We may finally assume also that f is represented in
B0 [t], hence represented in Ã0 [t]. Since Ã0 → Ã0 we are reduced to proving i
for Ã0 , where A0 is nötherian and regular. Then Ã0 is also nötherian by 3.4.1
and regular (since it has a regular completion). So we assume A given satisfying
3.4.1, and A nötherian and regular. Let f be monic in A[t]. We may assume f
59
3.4 HENSEL RINGS
irreducible and separable over the field K of fractions of A, and we have to show
A[t]/f is a local ring. Let L/K be a Galois extension in which f splits, and B the
normalization of A in L. Write A[t]/f ⊂ B. It suffices to show B is local. Let P
be a prime of B above p, and let B0 be the decomposition ring of B with respect
to the operation of G(L/K), p0 the maximal ideal of B0 under P. Then B0p0
is étale over A, with no residue field extension ([2] V, Proposition 2.2). Hence
B0p0 = A, hence B0 = A. But this means P is the only prime of B over m.
Q. E. D.
Corollary 3.4.3.
i If A is a hensel ring and a is an ideal of A then A/a is a hensel ring.
ii If A is a hensel ring and B/A is finite and local then B is a hensel ring.
iii If A is any local ring then Ã is a hensel ring.
For i, ii, iii apply i, ii, iii respectively of 3.4.2.
Definition 3.4.1. Ã is the henselization of A.
The map A → Ã is universal with respect to maps into hensel rings. As a

functor of A, Ã has very good properties, better in fact than the completion. For
instance, supposing A nötherian, we have
• A regular iff. Ã regular
• A reduced iff. Ã reduced
• A normal iff. Ã normal
These properties, and others, are easily deduced from [2] I, Section 9.
Theorem 3.4.4. Let A be a hensel ring. The category of finite and finitely
presented, étale A-algebras is equivalent with the category of finite separable
A-algebras. (Ā = A/mA).
60
Proof : The equivalence is of course given by B Ã B̄ = B/mB. To show that

every separable A-algebra L is isomorphic to a suitable B̄, we may assume L is a
separable field extension, so L ≈ Ā[t]/f¯ for some monic f¯. Let f ∈ A[t] (monic)
represent f¯, and set B = A[t]/f , so that B̄ ≈ L. Since (f¯0 ) = 1 in L, we have
(f 0 ) = 1 in B by Nakayama. Hence B/A is étale, finite, finitely presented, as
desired.
We have to show that the functor is bijective on Hom. Let B, C be given
/A. We may assume spec C connected, and it follows from [2] I, Corollary 5.4
that HomA (B, C) ⊂ HomA (B̄, C̄). Suppose given a map B̄ → C̄ and consider
the graph B ⊗Ā C → C̄. Since B̄, C̄ are separable over Ā, B ⊗Ā C ≈ C × L for
some L in such a way that the graph is the projection (spec C̄ connected by 3.4.2
ii). Since B ⊗A C is finite over A, it follows that B ⊗A C ≈ C 0 × D for some C’,
D with C̄ 0 ≈ C̄, D̄ ≈ L. Combining the projection with the map C → B ⊗A C
we get a map C → C 0 , and the induced map C̄ → C̄ 0 is an isomorphism. We
have to show C → C 0 is an isomorphism. But C’ is finite, finitely presented, and
flat, hence free. So a set of elements of C’ which is a basis mod m of C̄ 0 as an
Ā-module is a basis for C’ as an A-module. Therefore the images in C’ of a basis
∼
for C is a basis for C’, so C −→ C 0 .
Q. E. D.
Let G be a profinite group and denote by TG the topology of finite continuous

G-sets (cf. 1.1.5). Let X be a prescheme and f : TG → TX a morphism. Since the
category of TG - sheaves is equivalent with the category of continuous G-modules,
we may identify f S F , F ∈ SX , with a certain G-module. The underlying groups
of this G-module is easily seen to be lim F (f (Ḡ)) where Ḡ runs over the category
−→
Ḡ
of quotients of G with respect to open normal subgroups. Since the functor
associating with a G-module its underlying group is exact, it is clear that the
underlying group of the module corresponding to Rq f S F is
Rq lim F (f (Ḡ)) = lim H q (TX , f (Ḡ)); F

−→ −→
Ḡ Ḡ
So 2.4.2 reads (where e ∈ TG is the set of one element)
61
3.4 HENSEL RINGS
Spectral Sequence 3.4.1 (Hochschild-Serre).
E2p,q = H p (G, lim H q (TX , f (Ḡ); F )) =⇒ H ∗ (TX , f (e); F )

−→
Ḡ
Here the G-cohomology is the Tate cohomology. For instance, G might be

π1 (X), or a quotient by a closed normal subgroup of π1 (X). In such cases one
can often apply Corollary 3.3.7 to interpret the limit. Thus if X is quasi-compact
and quasi-separated, X given over a field k, and if X̄ = X ⊗k k̄ (k̄ the separable
algebraic closure of k), and π : X̄ → X is the map, one gets a spectral sequence
H p (G(k̄/k), H q (X̄, π ∗ F )) =⇒ H ∗ (X; F )
Let again A be a hensel ring and set k = A/mA, G = G(k̄/k). Theorem

3.4.4 and Galois theory yield a morphism f : TG → TX where X = spec A and
the topology in case (f).
Theorem 3.4.5. Let A be a hensel ring, X = specA, k = A/mA, F ∈ SX . With

the above notation
H p (G(k̄/k), f S F ) ≈ H p (X, F )
In particular, if k is separably algebraically closed, then H p (X; F ) = 0 for all

F ∈ SX , p > 0.
Proof : By 3.4.1, we have to show
Rq f S F = lim H q (TX , f (Ḡ); F ) = 0, q>0

−→
Ḡ
Set ←
X
−=←
lim f (Ḡ). ←
X
− is affine with ring −
A = lim A(Ḡ), A(Ḡ) the affine ring of
→
− −→
Ḡ
f (Ḡ). Each ring A(Ḡ) corresponds to a finite Galois extension k̄ ⊃ k(Ḡ) ⊃ k
and hence is connected, therefore local by 3.4.2 ii and hensel by 3.4.3 ii. It is
clear that therefore −
A is local and hensel, with residue field k̄. Let −
→ F be the
→
q S q
sheaf induced on ←X− by F. By Corollary 3.3.7, R f F ≈ H (X −; −
← F ). Hence we
→
are reduced to the case k = k̄ separably algebraically closed.
62
Let U/X be étale, finitely presented, and suppose U covers the closed point
of X (so that U covers X). Let B be the local ring of U at a point above the closed
point. Then by [2] I, Theorem 7.6 and 3.4.2 ii, B is étale, finite, and fin. pres..
Since k = k̄, B ≈ A by 3.4.4. Therefore X is initial in the category of U TX which
cover X. To show H q (X, F ) = 0, consider the identity map id : X → X. Clearly
Rq idF = 0, q > 0. Hence the stalk of Rq id∗ F is zero at every geometric point.
Let P → X be a geometric point of X above the closed point and apply Corollary
2.4.4 and Proposition 3.1.7. Since X is initial in the category of U covering X, we
find (Rq id∗ F )p ≈ H q (X, F ), therefore H q (X, F ) = 0, q > 0.
Q. E. D.
Lemma 3.4.6. Let X be a prescheme and ² : P → X a geometric point. Let I be

the category of pairs (U, φ); U ∈ TX , φ : P → U over ² in the case (1) topology.
The subcategory I 0 of I of pairs (U, φ) with U affine is initial in I, and lim U
←−
(U,φ)∈I 0
is the spectrum of a hensel ring with separably algebraically closed residue field
(the residue field is the separable algebraic closure of k(x) if x ∈ X is the center
of P).
Routine.
Corollary 3.4.7. Let π : Y → X be a finite morphism. Assume case (1). Then

Rq π∗ F = 0 for all F ∈ SY , q > 0.
Proof : We have to show the stalks are zero. Let ² : P → X be a geometric point.
By 2.4.4 and Proposition 3.1.7,
(Rq π∗ F )p ≈ lim H q (TY , U ×X Y ; F )

−→
(U,φ)∈I
≈ lim H q (TY , U ×X Y ; F )
−→ 0
(U,φ)∈I
where I, I 0 are as in 3.4.6. Since π is finite, it is affine, and therefore U ×X Y

is affine for all (U, φ) ∈ I 0 . Set spec A = lim U ; spec B = lim U ×X Y .We have
−→ −→
I0 I0
q q
(R π∗ F )p ≈ H (spec B, − F ) by Corollary 3.3.7 for suitable −
→ →F . B/A is finite,
63
3.5 COHOMOLOGICAL DIMENSION
and so B is a direct product of hensel rings by 3.4.6, 3.4.2 ii, 3.4.3 ii, and the
hensel rings have separably algebraically closed residue fields. Since cohomology
obviously commutes with direct sums of schemes, H q (spec B, −
F ) = 0, q > 0 by
→
3.4.5, so we are done.
Q. E. D.
3.5 Cohomological dimension

Let T be a topology, and F a presheaf of abelian groups on T. F is called a
torsion presheaf iff. F (U ) is a torsion group for all U ∈ T .
Proposition 3.5.1. Let f : T 0 → T be a morphism of topologies and F a torsion

sheaf on T. Suppose T is nötherian (cf. Section 2.5). Then Rq f S F is a torsion
sheaf on T 0 for all q ≥ 0.
Proof :Clearly, an inductive limit of torsion groups is a torsion group. Recalling

(Section 2.1) the construction of the functor ], we are reduced to showing Rq F
is a torsion presheaf where Rq F is as in 2.4.4, i.e., we are reduced to showing
H q (T, U ; F ) is a torsion group for all q ≥ 0, U ∈ T . Write
F = sup Fn
n∈N
where Fn is the kernel of multiplication by n in F (so that Fn is the subsheaf of

F of sections whose order divides n). By 2.5.3 we are reduced to the case F =
n
Fn , since T is nötherian. But the multiplication F → F induces multiplication
n
H q (T, U ; F ) → H q (T, U ; F ), as is seen by multiplying an injective resolution of
F by n. Since for F = Fn the multiplication is the zero map, it follows that
H q (T, U ; F ) is annihilated by n.
Q. E. D.
Corollary 3.5.2. Let π : X → Y be a morphism of preschemes, X and Y quasi-

compact and quasi-separated, and let F be a torsion sheaf on X. Then Rq π∗ F is
a torsion sheaf on Y, all q ≥ 0.
64
This is trivial in case (f) because then TX is nötherian. For the other cases,
one has to verify (notation of Section 3.1) that a sheaf is torsion iff. αS F (resp.
β S F , resp. γ S F ) is torsion, under the assumption (by 1.2.1) that these functors
are equivalences of categories. The ”only if” is obvious. To show ”if” we need,
∼
since αS αS F −→ F (resp. · · · ) to show that if G is a torsion sheaf then αS G is
a torsion sheaf. But the construction of αS G (cf. Section 2.4, 1.3.1, 2.1.1) is by
means of inductive limits and so it is clear.
Let X be a prescheme, F a sheaf on X, and i : x → X a point of X, where
x = spec k(x) is given the structure of scheme. F is said to be zero at x iff. i∗ F
is the zero sheaf. Obviously, if ² : P → X is a geometric point centered at x then
the stalk (cf. 1.2.6) of F at P is isomorphic to the stalk of i∗ F at ”P”. One sees
easily that F is zero at x iff. the stalk of F at P is zero for any geometric point
² : P → X centered at x.
Suppose X defined over a field k. For a point x of X, write dim x =
tr. deg.k k(x) (so dim x is the reference to k). We say a sheaf F on X is zero
in dimension > r iff. F is zero at every point x of X with dim x > r.
Definition 3.5.1. X has cohomological dimension ≤ d (write cd X ≤ d) iff.

H q (X, F ) = 0 for all torsion sheaves F and all q > d.
Theorem 3.5.3. Let X/k be a nötherian integral prescheme, k a separably al-

gebraically closed field. Assume the field R(X) of rational functions of X has
transcendence degree n over k. If F is a torsion sheaf on X, zero in dimension
> r, then H q (X; F ) = 0 for q > 2r. In particular, cd X ≤ 2n.
Proof : We may assume case (f). The assertion is trivial for r = −1 because a
sheaf which is zero at all points is the zero sheaf (apply Proposition 3.1.6 to all
U ∈ TX ). Moreover the theorem is constant for r ≥ n. We use induction on r.
Assume the theorem true for r = s − 1 and let F be a torsion sheaf on X, zero in
dimension > s. Let {iν : xν → X} be the set of points of X of dimension s, and
let Xν = x¯ν be the closure of xν with its canonical reduced structure. Xν satisfies
the conditions of the theorem if s is substituted for n, and and the induction
assumption implies that the theorem is true for Xν with r < s. Consider the map
65
F → iν∗ i∗ν F . Since X is nötherian and F is zero in dimension > s it is easy to see
that a section ξ ∈ F (U ) is mapped to zero in iν∗ i∗ν F for all but a finite number of
L
ν. Hence one obtains a map 0 → K → F → iν∗ i∗ν F → C → 0 (where X and
ν
C are the kernel and cokernel respectively). All these sheaves are torsion, and
F, ⊕iν∗ i∗ν F have the same stalks in dimension > s − 1. Therefore K and C are
zero in dimension > s − 1. Applying the induction assumption and examining
the relations in the cohomology, one reduces to the case F = ⊕iν∗ i∗ν F , hence by
2.5.4 to the case F = iν∗ i∗ν F . This sheaf is zero outside of X. Replacing Xν by X
and s by n we are reduced to the case F = i∗ F0 where i : x → X is the generic
point of X and F0 is a torsion sheaf on x = spec R(X). Moreover we may assume
the theorem true for r < n.
Lemma 3.5.4. Rq i∗ F0 is zero in dimension > n − q.
Assume the lemma, and examine the spectral sequence (cf. 2.4.2) E2p,q =
H p (X; Rq i∗ F0 ) =⇒ H ∗ (X; F0 ). We are interested in the terms E2p,0 = H p (X; i∗ F0 ).
Now cohomology over a field is equivalent with Galois cohomology, so we can ap-
ply the dimension theory for Galois cohomology to the ending H q (X; F0 ) of the
spectral sequence. Since tr. deg.k R(X) = n, it follows that H q (X, F0 ) = 0 for
q > n (cf. Sem. Bourb. n◦ 189, p. 12, Remarque). On the other hand, we are
assuming the theorem for r < n, and so it follows from the lemma and Proposi-
tion 3.5.1 that E2p,q = 0 for p > 2(n − q). The theorem can now be obtained by
examination of the spectral sequence.
Q. E. D.
Proof of 3.5.4: Let y ∈ X be given, dim y = m > n − q and let ² : P → X be a

geometric point centered at y. We have to show the stalk (Rq i∗ F0 )P is zero. Let
A be the hensel ring obtained in Lemma 3.4.6 and let A1 , · · · , Az be the affine
rings of the irreducible components of spec A. The rings Aj are hensel rings by
3.4.3 i. Let Rj be the field of quotients of Aj (A is reduced). Rj is immediately
seen to be (separably) algebraic over R(X), and, applying Corollary 3.3.7 (or well
L q
known facts about Galois cohomology) one finds (Rq i∗ F0 )P ≈ H (spec Rj ; Fj )
j
66
where Fj is some (torsion) sheaf induced by F0 . Since Rj is algebraic over R(X),

tr. deg.k Rj = n. Let K be the common residue field of A, Aj , so that K is the
separable algebraic closure of k(y). We have tr. deg.k K = m. Now an equi-
characteristic hensel ring contains fields over which the residue field is purely
inseparable (this is easy to see). Hence Rj contains a separably algebraically
closed field Kj with tr. deg.k Kj = m. So tr. deg.Kj Rj = n − m. The lemma now
follows from the dimension theory for Galois cohomology.
Q. E. D.
Corollary 3.5.5. Let X/k be a nötherian prescheme, k a field, and let X1 , · · · , Xr

be the irreducible components of X with their reduced structure. Then
cd X ≤ cd k + 2 max{tr. deg.k R(Xj )}
Applying dimension theory of Galois cohomology and spectral sequence 3.4.2

one reduces to the case k separably algebraically closed. Then the corollary follows
Lr
by induction from 3.5.3 and the map F → Fi where Fi is the sheaf induced on
i
Xi (cf. also [2] I, Theorem 8.3).
67
68
4 CALCULATIONS FOR CURVES AND SURFACES
4 Calculations for curves and surfaces
4.1 General facts
We assume throughout this chapter that we are in case (1) or case (f) and
that X is quasi compact and quasi separated in case (f).
Let X → Z be a morphism of preschemes, and A a group scheme over Z.
Because of 3.1.4, the functor HomZ ( , A): TX → Ab is a sheaf on X. We will
denote such a sheaf by AX , or merely A if it will not cause confusion. Of course,
AX ≈ HomX ( , A ×Z X). It follows immediately from 3.3.5, if A is separated and
finitely presented over Z, that under the assumptions of Corollary 3.3.7 the sheaf
HomZ ( , A) commutes with the limit, i.e., that, denoting by Fi the sheaf induced
by AX on Xi (Fi ≈ AXi ) we have − F ≈ A X (notation of Corollary 3.3.7). This is
→ ←−
also true in the case A is discrete (below).
Case A is discrete: An abelian group A determines canonically a group
`
scheme over spec Z, namely spec Z with the obvious group law. We denote
|A|
this scheme also by A, and hence have defined AX . This sheaf can also be obtained
in other ways. In fact, AX is isomorphic to the associated sheaf of the constant
presheaf whose group of sections on each U ∈ TX is A. Therefore we call AX
a constant sheaf . Or, denoting by i : {X} → TX the inclusion of the discrete
topology {X} (cf. 2.2.1), AX ≈ iS A = (iP A)] . In particular, the notation ZX
does not conflict with 2.1.1 (however Z 6= spec Z). Finally, suppose U ∈ TX
consists of finitely many (say r) connected components. Then AX (U ) ≈ Ar .
If X is nötherian, integral, and normal, and if i : x → X is the generic point,
then i∗ Ax ≈ AX . This follows easily from [2] I, Prop. 10.1. But, it is false in
general. For instance, for a nodal rational curve X, one gets 0 → AX → i∗ Ax →
AQ → 0 where AQ is the constant sheaf at the node Q (extended by zero).
Among non-discrete group schemes, we will be particularly interested in the
sheaf (Gm )X of units of Γ(U, OU ) (U ∈ TX ). (Here Gm = spec Z[t, 1/t]), and the
69
4.1 GENERAL FACTS
sheaf of n-th roots of unity µn . Suppose n is prime to char k(x) for all points
n
x ∈ X. Then the n-th power map Gm → Gm yields
n
0 → µnX → GmX → GmX → 0 (n prime to the residue char.) (4.1.1)
In fact, this sequence is always left exact, and under the assumptions, extraction
of an n-th root of a unit gives an étale covering. Hence the cokernel is zero, as a
sheaf.
Proposition 4.1.1. H 1 (X, Gm ) ≈ Pic X.
Proof : Assume first X affine. Then we may assume we are in case (f). By
2.3.1. H 1 may be computed as Čech cohomology. Since for a covering {Uα→X }
`
in case (f) Uα → X is faithfully flat and quasi-compact, the result follows from
descent theory (cf. Sem. Bourb. n◦ 190. 4. e) and the fact the topology ocntains
sufficiently fine Zariski open coverings. For the general case, let T Z be the Zariski
topology on X, T G the Grothendieck topolgy (case (1)), and f : T Z → T G the
inclusion. Knowing the result for affines, one finds R1 f S Gm = 0, and so we are
done by the Leray spectral sequence 2.4.2.
Q. E. D.
Suppose now X is integral, and let i : x → X be the generic point (x =

spec R(X), R(X) the field of rational function of X). We have H 1 (x, Gm ) = 0
by 4.1.1 (this is ”Hilbert theorem 90”). More generally, H 1 (U, Gm ) = 0 for all
U ∈ TX . Hence R1 i∗ (Gm )X = R2 i∗ (Gm )X = 0. By 2.4.2 we find
H 1 (X, i∗ Gm ) = 0 (i : x → X the gen. pt., X integral) (4.1.2)
There is a natural inclusion (Gm )X ,→ i∗ (Gm )x . We write
0 → Gm → i∗ Gm → DX → 0 (i : x → X the gen. pt., X integral) (4.1.3)
Here DX is the sheaf of ”Cartier divisor” on X. Combining 4.1.2, 4.1.3 one gets
R(X) → H 0 (X, DX ) → Pic X → 0 (X integral) (4.1.4)
70
It is clear from the ”five lemma” that H 0 (X, DX ) is the same as it would be if
DX were defined by 4.1.3 in the Zariski topology (because H 1 (X, i∗ Gm ) = 0 also
in the Zariski topology).
If X is moreover nötherian and regular, DX is easily identified as
M
DX ≈ iν∗ Zxν (X nöth. reg.) (4.1.5)
ν
where {iν : xν → X} are the generic points of the irreducible closed subsets
of codim 1. Since in the Tate cohomology H 1 (G, Z) = 0 for any G one finds
H 1 (xν , Z) = R1 iν∗ Z = 0. Therefore by the Leray spectral sequence and 2.5.4 (TX
is nötherian)
H 1 (X, DX ) = 0 (X nöth. reg.) (4.1.6)
We get from 4.1.3 and the fact that R1 i∗ (Gm )x = 0,
H 2 (X, Gm ) ,→ H 2 (x, Gm ) (X nöth. reg.) (4.1.7)
Presumably, if X is affine, H 2 (X, Gm ) is closely related to the Brauer group of its

ring (cf. [6]).
4.2 Case of an algebraic curve
Let X be a nötherian integral scheme over a separably algebraically closed

field k0 and suppose the field R(X) of rational functions of X has transcendence
degree 1 over k0 . In particular, X could be an algebraic curve. Let i : x → X
be the generic point. We have H q (x, Gm ) = 0 for q ≥ 2 by dimension theory
for Galois cohomology, for q = 1 by 4.1.1 and for q = 2 by Tsen’s theorem (cf.
[10]) if k0 is algebraically closed. In general, H 2 (x, Gm ) is anyhow a p-group
(p = char k0 ) as is seen from 4.1.1. This being true for any algebraic extension
of R(X), one finds Rq i∗ Gm are p-groups for q > 0, and hence
H q (X, i∗ Gm ) = 0 (ignoring p-torsion) (4.2.1)
71
4.2 CASE OF AN ALGEBRAIC CURVE
If X is a complete algebraic curve, so that H 0 (X, Gm ) ≈ k0∗ (which is divisible by

n if (n, p) = 1), one finds from 4.1.1


 µn q=0


 (Pic X) q=1
n
H q (X, µn ) = (X complete, (n, p) = 1) (4.2.2)

 (Pic X/n) ≈ Z/n q=2



0 q>2
where (Pic X)n is the group of points of order n on the Jacobian of X. These are
the expected values (However, one cannot hope for ”good” values of H q (X, Z)
since for instance H 1 (X, Z) = 0 if X is regular). Note that in 4.2.2 the removal
of a point of X has the effect of killing the H 2 .
If X is regular, so DX torsion free, one deduces from 4.1.1 and 4.1.3
0 → µn → i∗ Gm → i∗ Gm → DX /nDX → 0 ((n, p) = 1) (4.2.3)
which is an acyclic resolution of µn (possible p-torsion does not matter). So one

may write
H 1 (X, µn ) = {f ∈ R(X)∗ |(f ) ∼

= 0(n)}/{f ∈ (R(X)∗ )n } ((n, p) = 1)
(4.2.4)
where (f) denotes the divisor of f.
Let again i : x → X be the generic point of X, and F any sheaf on x. Clearly
q
R i∗ F is zero at x (q > 0), hence its cohomology vanishes in dimension > 0 by
3.5.3. Therefore the Leray spectral sequence H p (X, Rq i∗ F ) =⇒ H ∗ (x, F ) reduces
to an exact sequence
· · · → H q (X, i∗ F ) → H q (x, F ) → H 0 (X, Rq i∗ F ) → · · · (4.2.5)
In fact, by the dimension theory of Galois cohomology, H p (x, F ) and Rq i∗ F are

¯
zero for q > 2. If F is torsion or corresponds to a divisible G(R(X)/R(X))-
module, they are zero for q > 1. The relevant part of 4.2.5 is then
0 → H 1 (X, i∗ F ) → H 1 (x, F ) → H 0 R1 → H 2 (X, i∗ F ) → 0 (4.2.6)

L
Suppose X is regular. Then H 0 R1 ≈ H 1 (xp , Fp ) where p runs over closed
p
points of X. xp is the generic point of the henselization of OX,p and Fp is the
72
induced sheaf of xp . This is essentially the situation studied by [12] in the case
F = Ax is points with values in an abelian variety A/x. H 1 (X, i∗ Ax ) is then
the group of ”locally trivial principal homogeneous spaces” of A. Ogg derives a
kind of duality between H 0 (X, i∗ Â) and H 2 (X, i∗ A) (Â the dual abelian variety).
By similar methods one may obtain a perfect duality between H q (X, i∗ F ) and
H 2−q (X, i∗ F̂ ) when F is points with values in a finite group scheme over x of
dimension (as R(X)-module) prime to p and F̂ is its Cartier dual. The duality is
given by a cup product into H 2 (X, µ) ≈ Q/Z (prime to p), µ the sheaf of roots
of unity.
4.3 Local calculations in dimension 2

Let k0 be a separably algebraically closed field. We denote by k0 {x, y, · · · , z}
the henselization of the local ring at (0, · · · , 0) of spec k0 [x, y, · · · , z]. It is known
that k0 {x, y, · · · , z} consists of those power series α ∈ k0 [[x, y, · · · , z]] which
are algebraic over k[x, y, · · · , z]. We will generally denote the closed point of
spec k0 {x, y, · · · , z} by Q.
In the case of one variable k0 {t}, set k = k0 ({t})=field of fractions of k0 {t}.
The only algebraic extension of k of degree prime to p = char k0 is k[t1/n ], i.e., if
G = G(k̄/k) is the Galois group of the separable algebraic closure, and if Ḡ is the
Galois group of the maximal extension prime to p, one has G ≈ (Ẑ)(prime to p) =
inverse limit of Z/n, (n, p) = 1, and the kernel of the map G → Ḡ is a p-group.
(This isomorphism is not canonical). One finds


 µn
 q=0
q q
H (spec k, µn ) = H (Ḡ, µn ) ≈ Z/n q = 1 ((n, p) = 1) (4.3.1)


 0 q>1
Here spec k = (spec k0 {t}) − Q.

Consider the case of two variables k0 {t, x}, and let
R = (k0 {t, x})t = k0 {t, x}[1/t]
We wish to calculate the cohomology of spec R = spec k0 {t, x}- locus of t = 0.
73
4.3 LOCAL CALCULATIONS IN DIMENSION 2
Now R ⊃ k = k0 ({t}). Set
R̄ = R ⊗k k̄ = lim R ⊗k k 0
−→0 k
when k̄ is the separable algebraic closure of k, and k’ runs over finite subextensions
of k̄. Choose such a k’ and let t’ be a local parameter of the normalization o0
of k0 {t} in k’, so that o0 ≈ k0 {t0 }. Then it is easily seen that R0 = R ⊗k k 0 ≈
(k0 {t0 , x})t0 which is a ring similar to R. Now R (hence R’) is integrally closed.
Moreover every prime of R is maximal. Finally, R is nötherian and Pic R = 0.
Hence R is a PID. Since R̄ is a limit of such rings, Pic R̄ = 0 and R̄ is integrally
closed. Since R̄ has transc. deg. 1 over k̄, every prime of R is maximal. Actually,
R̄ is also nötherian, hence a PID. To show this one has to show every prime
ideal finitely generated. This is equivalent to showing every prime is induced
from some R’, i.e., to showing the primes of R’ split into finitely many primes in
R̄, which amounts to showing R/p finite over k for all primes p of R. Now p is
induced by some prime element f ∈ k0 {t, x}, t not dividing f. Write k0 {t, x} as a
limit of rings étale, localized from an algebra of finite type over k0 {t}[x]. We can
find f in some such ring, say A. Then A/(f ) is of finite type over k0 {t}. Applying
ZMT, A/(f ) is localized from a finite k0 {t}-algebra. Hence A/(f ) is finite over
k0 {t} and hensel by 3.4.2 ii and 3.4.3 ii. Therefore A/(f ) ≈ k0 {t, x}/(f ) and we
are done.
By 4.1.1 and 4.2.1
H q (spec R̄, Gm ) = 0, q > 0 (ignore p) (4.3.2)
Now in a hensel ring over a separably algebraically closed field, the group of units
is obviously divisible by n, (n, p) = 1. Hence for an R’ as above, R0∗ /(R0∗ )n ≈ Z/n,
the generator being the residue of t’. Since t1/n is in R̄, R̄∗ is divisible by n.
Therefore 4.1.1 and 4.3.2 yield
H q (spec R̄, µn ) = 0, q > 0 ((n, p) = 1) (4.3.3)
Applying the Hochschild-Serre spectral sequence 3.4.2:
H q (spec R, Gm ) = 0, q > 0 (ignore p) (4.3.4)
74


 µn
 q=0
H q (spec R, µn ) = Z/n q = 1 ((n, p) = 1)


 0 q>1
Set Y = spec k0 {t, x} − Q (Q the closed point) and j : spec R → Y the

map. We have spec R = Y − X where X = spec k0 ({t}) is the locus t = 0.One
sees easily that Rq j∗ Gm = 0, q > 0 (ignore p). There is an exact sequence
0 → GmY → j∗ Gm → ZX → 0, and hence H q (Y, Gm ) ≈ H q−1 (X, Z) by 4.3.4, for
q > 1. Clearly H 1 (Y, Gm ) = 0. We have H q−1 (X, Z) ≈ H q−2 (X, Q/Z) because
of the sequence 0 → Z → Q → Q/Z → 0, and so H 2 (X, Z) ≈ Q/Z ˆ (prime to p).
Therefore


 µn
 q=0
H q (Y, µn ) ≈ 0 q = 1, 2 ((n, p) = 1) (4.3.5)


 Z/n q = 3
This is the expected result if one thinks of Y as a 4 ball minus point.
We will also need to know the cohomology of spec S where S = (k0 {x, y})xy =
k0 {x, y}[1/xy]. This could be obtained from 4.3.4 but for our purposes the
following method is better. Set t = xy, so that S ⊃ k = k0 ({t}), and set
S̄ = S ⊗k k̄ = lim S ⊗k k 0 (k 0 /k finite separable). Here S ⊗k k 0 is not localized
−→
k0
from a regular ring, but one can show anyhow that Pic S ⊗k k 0 is zero (cf. Remark
3). S̄ is seen to be a PID by a similar argument to that used for R̄ above, and
therefore H q (spec S̄, Gm ) = 0, q > 0 (ignore p). However, H 0 (spec S̄, Gm ) = S̄ ∗
is not divisible. In fact, computing a little, one finds S̄ ∗ /(S̄ ∗ )n ≈ (Z/n ⊕ Z/n)/∆
(∆ ”diagonal”, (n, p) = 1) with canonical generators x Ã (1, 0) and y Ã (0, 1).
Hence H 1 (spec S̄, µn ) = (Z/n ⊕ Z/n)/∆. x and y could also be viewed as gener-
ators of this group when written in the form 4.2.4. Applying 3.4.2,


 µn
 q=0
H q (spec S, µn ) = Z/n ⊕ Z/n q=1 ((n, p) = 1)


 Hom(G, (Z/n ⊕ Z/n)/∆) q = 2
(4.3.6)
75
4.4 DIGRESSION ON THE PICARD GROUP OF A CURVE OVER K0 {T }
4.4 Digression on the Picard group of a curve over k0 {t}

Write o = k0 {t} (k0 algebraically closed). In this section, V denotes a scheme
proper over spec o, which is irreducible and nonsingular, of dimension 2. We
denote by U the generic fibre of V, and we assume U is geometrically simple over
k = k0 ({t}) (U is an algebraic curve). Finally, we assume V /spec o has a section.
In this situation, if X denotes the closed fibre, then U is the open subscheme V-X
of V. Let Xi (i = 1, · · · , m) be the irreducible reduced components of X, and say
P
X = ri Xi (cf. 4.4.1 e).
The scheme V /spec o, being of finite type, can be obtained by a base exten-
sion from some V0 /spec o0 where o0 is a geometric discrete valuation ring over
k0 and o/o0 is étale, and V0 may be obtained by localization from an algebraic
surface defined over k0 . The following properties, listed for the convenience of
the reader, may be obtained with no difficulty from classical surface theory by
applying such a descent, if they are not obvious anyhow.
Claim 4.4.1.
a Resolution of singularities if the assumption that V is nonsingular is dropped.
b V is projective over spec o.
c Intersection theory for (Cartier) division on V, of the form (D.Y ) ( an

integer) where D is any divisor and Y is a divisor with support on the
closed fibre X. With this restriction, (D.Y ) depends only on the divisor
class of D. Hence the symbol may also be used if D denotes a divisor class.
d There exists a divisor class K, called canonical, appearing in the genus

formula e.
P
e Let Y = si Xi be a divisor with support on X, and assume Y > 0. One
associates in the obvious way a closed subscheme of V to Y, which we denote
by the same letter. The following formular holds:
1
1 − χ(Y ) = ((Y 2 ) + (X.Y )) + 1 , p(Y ) (χ the Euler Char.)
2
The definition of p(Y ) is extended formally to nonpositive Y.
76
f Castelnuovo’s criterion for exceptional curves (of the first kind): p(E) = 0,
(E 2 ) = −1 (E reduced, irreducible, with support on X), and the factoriza-
tion of regular birational maps into locally quadratic transformations.
g If Y has support on X and (Y 2 ) = 0 then Y = aX for some integer s

(This follows from Hodge’s theorem on a surface, and the fact that V has
a section)
Proposition 4.4.2. Consider the canonical map Pic V → Pic X.
i This map is surjective.
ii ker(Pic V → Pic X) = K is uniquely divisible by n prime to char k0 = p.
We omit the proof.

Denote by Pic0 X the subgroup of Pic X consisting of divisor classes whose
degree on each reduced, irreducible component Xi of X is zero, and by Pic0 V
the inverse image of Pic0 X under the map Pic V → Pic X. Pic0 V is the group
of divisor classes D on V such that (D · Xi ) = 0 for al components Xi of X.
Now Pic0 X is easily seen to be divisible by n prime to p. Since obviously K =
ker (Pic0 V → Pic0 X) one finds by 4.4.2
Corollary 4.4.3.
i Pic0 V is divisible by n.
ii (Pic V )n = (Pic0 V )n ≈ (Pic0 X)n = (Pic X)n ((n, p) = 1), where ( )n

denotes the subgroup of elements of order n.
P
Let D be the group of divisors (of the form si Xi with support on X),
and D̄ its image in Pic V . By 4.4.1 g, D̄ ≈ D/(X) where (X) is the subgroup
generated by X. Clearly, D is the kernel of the map Pic V → Pic U . This map is
surjective because V is nonsingular. Let A/spec k be the Jacobian of U, and for
a field k 0 /k denote by Ak0 the points of A with values in k’. Ak ⊂ Pic U is the
subgroup of divisor classes of U of degree zero. Now by 4.4.1 g, D̄ ∩ Pic0 V = 0.
Hence the map Pic V → Pic U induces an injection Pic0 V ,→ Pic U . Clearly its
image is in Ak , and we denote this image by A0k .
77
Lemma 4.4.4. A0k depends only on U/k.
Proof : Given a curve geometrically simple, proper over k, one cam imbed U as
an open subscheme of some V /spec o as above, using 4.4.1 a. Applying 4.4.1 a
∼
and f. one reduces to showing Pic0 V → Pic0 V 0 when f : V 0 → V is obtained by
blowing up a point P of V. We have to show that if D’ is a divisor on V’ with
(D0 · Xi0 ) = 0 for all components Xi0 of X’, then D’ is linearly equivalent to f −1 (D)
for some D with (D · Xi ) = 0 all i. This is easy.
Q. E. D.
Proposition 4.4.5. Ak /A0k is a finite group.
Proof : Consider the diagram
/ Zm /²
DO O O
/ Pic V / Pic U
DO O O
/ / A0
0 Pic0 V k
where ² is the cokernel of A0k → Pic U . We claim the rows and columns are
exact if zeros are added on the periphery. Here m is the number of irreducible
components of X, and Pic V → Zm is the map D Ã ((D · X1 ), · · · , (D · Xm )).
This map may be factored as Pic V → Pic X → Zm where the second arrow is
the ”degree”, and is surjective. Clearly therefore the middle column is exact. The
only remaining one which is not obvious is the top row, but D̄ → Zm is injective
because of 4.4.1 g.
From the top row, since D̄ has rank (m − 1) over Z, ² is a group of finite type
and of rank 1. But Pic U/Ak ≈ Z is also of rank 1. Since we have 0 → Ak /A0k →
² → Pic U/Ak → 0, the proposition follows.
Q. E. D.
78
Note that since A0k is divisible if char k0 = 0, one can characterize A0k as the
maximal divisible subgroup of Ak in that case.
Let k 0 /k be a finite separable extension, and let o0 be the normalization of o
in k’, U 0 = U ⊗k k 0 . Constructing a suitable V 0 /spec o0 with generic fibre U’, one
defines A0k0 ⊂ Ak0 . By 4.4.1 a, V’ may be found with a map onto V (over o0 /o).
Therefore the map Ak ,→ Ak0 carries A0k into A0k0 . Set
[
A0k̄ = A0k0 (k̄ the sep. alg. clos. of k, k 0 /k finite)
k̄⊃k0 ⊃k
so that A0k̄ ⊂ Ak̄ . Denote by (Z)f the subgroup of an abelian group Z of elements
of finite order prime to p = char k0 .
Theorem 4.4.6. Suppose X reduced and with only ordinary nodes as singulari-
ties. Then
a (A0k̄ )f = (A0k )f (in other words, G(k̄/k) operates tivially on (A0k̄ )f )
b If a ∈ (Ak̄ )f and σ ∈ G(k̄/k) then σa − a ∈ A0k .
Proof : Let a ∈ (A0k̄ )f , and let k 0 /k be a finite separable extension of k in which a

becomes rational. Let o0 be the normalization of o in k’ and t’ a local parameter
of o0 . Then o0 = o[t0 ]. Set V 0 = V ⊗o o0 . V’ is easily seen to be nonsingular except
above the nodes of X. Above the nodes, V’ is anyhow complete intersection, hence
Cohen- Macauley, and therefore normal by Serre’s criterion since there are no
singular curves on V’. Again since V’ is Cohen-Macauley, one sees immediately
that the closed fibre X’ of V’ is reduced. The map X 0 → X is 1-1, hence an
isomorphism (because nodes are ”minimal singularities”).
Let V 00 → V 0 be obtained by resolving the singularity of V’, and choose
V” minimal with this property. Let X” be the closed fibre of V” and θ1 , · · · , θs
the connected components of the exceptional locus for the map V 00 → V 0 , say
θi = θi1 ∪ · · · ∪ θiri where the θiν are the irreducible components. Denoting by X 0
the proper transform on V” of the closed fibre X’ of V’, we have
X
X 00 = X 0 + aνi θiν
i,ν
79
some aνi > 0. We claim aνi = 1, all i, ν, and that the curves θi consist of a
configuration as depicted below (with suitable ordering of θiν )
(4.4.1)
>>} <<¡ ??£ AÄ
} >>> ¡¡<< ££?? ÄÄA
0 } > 1 ¡¡ << ££ ?? ri ÄÄ A
X } > θ ¡ <<< · · · £ ?θ Ä AX 0
}
>>i ¡¡
>¡ < £££ ??i ÄÄ
?
Ä
A
A
} ¡¡ >> << ££ ÄÄ ?? A
} ¡¡ > > << ££ ÄÄ ?? A
} ¡ £ Ä
all intersections being transversal and θiν nonsingular and rational with (θiν 2 ) =
−2.
Remark 3. This is of course purely a local statement, although the proof below
utilizes the whole scheme V”. The configuration may be used to prove Pic S̄ = 0
where S̄ is the ring used for 4.3.6.
These assertions may be verified by routine calculation: Note first that if

X 00 − Z > 0 (Z > 0) and if Z contains no exceptional curve then p(X 00 − Z) <
p(X 00 ). For,
1
p(X 00 − Z) = ((X 002 ) + (Z 2 ) − 2(X 00 · Z) − (K · X 00 ) − (K · Z)) + 1
2
1
= p(X 00 ) + ((Z 2 ) − (K · Z))
2
By 4.4.1 g and the fact that X” has components with multiplicity 1, (Z 2 ) < 0,
and (K · Z) > 0 because Z contains no exceptional curve.
Therefore, since V” was chosen to be minimal over V’,
X
p(X 00 ) > p(X 0 + θiν )
i,ν
with euqality iff. all aνi are equal 1. Calculating with 4.4.1 e,
P P P
p(X 0 + θiν ) = p(X 0 ) + p(θiν ) + (X 0 · θiν )
i,ν i,ν i,ν
P P P
s (4.4.2)
+ ( (θiν · θiµ )) − ri
i ν<µ i=1
P
Here p(θiν ) ≥ 0, (X 0 · θiν ) ≥ 2s since θi lies above a node of X’, and
i,ν
X
(θiν · θiµ ) ≥ ri − 1
ν<µ
80
since θi is connected. Moreover, p(X 0 ) ≥ p(X 0 ) − s since separation of a node

reduces p by 1 (s is at most equal to the number of nodes). So combining,
s
X s
X
00 0
p(X ) ≥ p(X ) − s + 2s + (ri − 1) − ri = p(X 0 )
i=1 i=1
But by the invariance of Euler characteristic under specialization, p(X 0 ) = p(U 0 )

if U 0 is the general fibre of V 0 . Since U 0 is also the general fibre of V 00 , we
have p(X 0 ) = p(X 00 ). Therefore all the above inequalities are equalities. Since
p(θiν ) = 0, (θiν 2 ) must be less than −1 by 4.4.1 f. We have (X 00 · θiν ) = 0 (since
X 00 is principal), hence (X 00 − θiν · θiν ) ≥ 2. But by the intersection numbers
appearing in 4.4.2, the average value of (X 00 − θiν · θiν ) for i, ν varying is 2. Hence
(X 00 − θiν · θiν ) = 2 for all i, ν and so (θiν 2 ) = −2. It is now easily checked that
4.4.1 is the only configuration possible.
Returning to the proof of 4.4.6 a, we had a ∈ (A0k̄ )f rational over k 0 . This
a corresponds to a unique divisor class D00 ∈ (Pic0 V 00 )f , so (D00 · · · Xj00 ) = 0
for all components Xj00 of X 00 . In particular (D00 · θiν ) = 0 for all i, ν. Because
the θiν are rational curves with (θiν 2 ) = −2, it is known that this implies the
transform D0 of D00 on V 0 is locally principal (cf. [7]). D0 of course has finite
order. Hence D0 induces an element d ∈ (Pic X 0 )f ≈ (Pic X)f . By 4.4.3, there
exists a D ∈ (Pic0 V )f inducing d. One finds the transform of D on V 00 is D00 , so
a is proved.
To prove b, let a ∈ (Ak̄ )f and choose k 0 /k as above, but a Galois extension,
so that a is rational over k 0 . Retain the above notations. We may find a divisor
D00 on V 00 inducing the class a in Pic U 00 = Ak0 . Say a is of order n. Then there
is a rational function f on V 00 such that
X
nD00 = (f ) + si Xi
some si (Xi00 the components of X 00 ). Now G = G(k̄/k) operates on V 0 /V by

automorphisms, and since V 00 was chosen minimal over V 0 , G also operates on
V 00 /V by automorphisms (such a minimal model is unique). The components of
X 00 lifting from components of X are certainly left fixed by G. But, examining
figure 4.4.1, it is also clear that the θiν s must be left fixed. Hence X 00 is left
81
4.5 COHOMOLOGY OF A CURVE OVER K0 {T }
componentwise fixed. Obviously σa (σ ∈ G) is induced by the transformed

divisor D00σ , and we have
X X
(nD00 )σ = n(D00σ ) = (f σ ) + si Xi00σ = (f σ ) + si Xi00
Hence
n(D00σ − D00 ) = (f σ ) − (f ) = (f σ /f )
so D00σ − D00 ∈ (Pic0 V 00 )n , and σa − a is therefore in A0k by part a.
Q. E. D.
4.5 Cohomology of a curve over k0 {t}

Let V /spec o, U, X be as in the previous section. We assume further that
X is reduced, and has only nodes as singularities, and that U has positive genus
(the case that U has genus zero is easier). We want to prove
∼
i H q (V, µn ) −→ H q (X, µn ) all q ((n, p) = 1)
(4.5.1)
ii H q (V, Gm ) = 0 q≥2 (ignoring p-torsion)
The map for i is canonical, and is induced by the map µnV → i∗ (µnX ), where
i:X →V.
Now H 2 (X, µn ) ≈ Pic X/n ≈ (Z/n)m (m= no. of comp. of X), and
H q (X, µn ) = 0 for q > 2. By 4.4.2, Pic X/n ≈ Pic V /n. Therefore, assum-
ing 4.5.1 i, the exact sequence 4.1.1 shows H q (V, Gm ) is torsion free, q ≥ 2, prime
to p. Hence 4.5.1 ii follows from 4.5.1 i once one knows H q (V, Gm ) is torsion for
q ≥ 2, and that fact can be checked easily from the exact sequence
0 → GmV → j∗ GmU → D(X) → 0
(j : U → V the injection and D(X) the sheaf of divisors on V with support on

X). So we concentrate on 4.5.1 i.
The isomorphism is obvious for q = 0. For q = 1, we have from 4.1.1
0 → o∗ /(o∗ )n → H 1 (V, µn ) → (Pic V )n → 0
82
Since o is a Hensel ring, o∗ is divisible by n. Hence H 1 (V, µn ) ≈ (Pic V )n .

Similarly, H 1 (X, µn ) ≈ (Pic X)n . So by 4.4.3 ii (and trivial functorality) the
assertion is true for q = 1. The case q = 2 is harder. We use the relative
cohomology sequence 3.2.7 for the map i : X → V , and 4.1.1, to obtain an exact
commutative diagram
1
HX (Gm )/n / H 2 (µ ) (4.5.2)
X n
² ²
0 / Pic V /n / H 2 (V, µ ) /² /0
n
² ²
0 / Pic U/n / H 2 (U, µ ) /δ /0
n
where ², δ are the cokernels. The top terms are easily calculated by using 3.2.3,
3.2.4, 3.2.5 and the spectral sequence H p (X, Rq i! F ) =⇒ HX
∗
(F ). One finds
1
HX (Gm ) ≈ Zm and HX
2
(µn ) ≈ (Z/n)m (m= no. of comp. of X). Hence the
top arrow of 4.5.2 is an isomorphism. Therefore ² → δ is an injection.
Now if V’ is an étale covering of V, and U’, X’ are the fibres of V’, it is clear
that the diagram 4.5.2 is mapped commutatively into the similar one for V’, U’,
X’.
Lemma 4.5.1. The image of H 2 (V, µn ) in H 2 (U, µn ) is annihilated in some étale

covering V’ of V.
Assume the lemma and consider the Hochschild-Serre spectral sequence 3.4.1
with G = π1 (V ). From the map µnV → i∗ (µnX ) one gets a morphism of spectral
sequences with ending H ∗ (V, µn ) → H ∗ (V, i∗ µn ) ≈ H ∗ (X, µn ). We have, in the
notation of 3.4.1, H q (TX , f (Ḡ); F ) ≈ H q (V 0 , F 0 ) by 2.4.7, setting V 0 = f (Ḡ) and
F’ the induced sheaf. Applying 4.5.1 to the diagram 4.5.2, one sees
lim H 2 (V 0 , µn ) ≈ lim Pic V 0 /n

−→ −→
Ḡ Ḡ
But by Proposition 4.4.2

∼
lim Pic V 0 /n −→ lim Pic X 0 /n
−→ −→
83
Taking into account the isomorphisms in dimension 0, 1 one finds
lim H q (V 0 , µn ) → lim H q (X 0 , µn ) for q = 0, 1, 2

−→ −→
(In fact, one gets 0 for q=1, and 0 for q=2 if X’ contains no rational curves.)
Therefore it follows from the spectral sequence that
∼
H q (V, µn ) −→ H q (X, µn ) for q = 0, 1, 2
It remains to show H q (V, µn ) = 0 for q > 2. (By 3.5.4 this is anyhow true if
q > 4.) For this purpose, examine the Leray spectral sequence
H p (V, Rq j∗ µn ) =⇒ H ∗ (U, µn ) (j : U → V ) (4.5.3)
One finds, using 4.1.1, R1 j∗ µn ≈ D(X)/n (D(X) the divisors with supports on
X). If π : X̄ → X is the normalization of X, it is easily seen that D(X)/n ≈
π∗ (Z/n)X̄ . Hence H p (V, R1 j∗ µn ) ≈ H p (X̄, Z/n) by 3.4.7. This is computable by
Section 4.2 since Z/n is isomorphic (not canonically) to µn . By 3.5.4, R2 j∗ µn is
concentrated at the closed points of X, and Rq j∗ µn is concentrated at the nodes
of X, and is computable there.
To calculate the ending H ∗ (U, µn ) of 4.5.3, use the spectral sequence 3.4.2.
The nonzero terms of E2p,q are (cf. 4.2.3)
Z/n H(G, Z/n)

E2p,q =E p,q
= (Ak )n H(G, (Ak̄ )n ) (G = G(k̄/k)) (4.5.4)
µn H(G, µn )
Because of the special nature of G (cf. Section 4.3) one has a duality for the G-
cohomology, due to [12] from which, together with the autoduality of the Jacobian
A, one finds the Euler characteristic
Y
χ(U, µn ) = (−1)q ](H q (U, µn )) = 0
q
(](Z)= no. of elts. of Z).
84
χ(U, µn ) can also be calculate from the E2p,q terms of 4.5.3 and the calculation
shows that the unknown terms ](H 3 (V, µn )) and ](H 4 (V, µn )) are equal. So we
are reduced to showing H 4 (V, µn ) = 0.
Since H 4 (U, µn ) = 0, one sees from 4.5.3 that the map H 2 R1 → H 0 R4 =
H 4 (V, µn ) is surjective, so it would be convenient if H 2 R1 were zero. To achieve
this, let V − be obtained from V /o by removing a section si passing through each
∼
component Xi of X. It is easy to see that H q (V, µn ) → H q (V − , µn ) for q ≥ 3
(apply 4.3.4 and 3.4.5 to the injection V − → V ). But in the Leray spectral
sequence similar to 4.5.3 for V − , the term H 2 R1 is zero, so we are done.
Proof of Lemma 4.5.1: To begin with note that the assumption that X be reduced
with only nodes is preserved in an étale covering V 0 /V . Set
Z = im(H 2 (V, µn ) → H 2 (U, µn ))
From 4.5.4 we have a sequence
0 → H 1 (G, (Ak̄ )n ) → H 2 (U, µn ) → Z/n → 0
where the Z/n is canonically isomorphic to Pic Ū /n. This Pic Ū /n will be anni-
hilated in a connected étale covering of degree n, hence in some V 0 /V if we know
Pic V contains elements D of order n (as usual by adjoining f 1/n for an f with
(f ) = nD). Such D are easily seen to exist by 4.4.3 ii since X is reduced with
only nodes and P (X) > 0. So for some V 0 /V the image of Z will be in the image
of H 1 (G, (A)n ), which is contained in H 1 (G, (A0 )n ) (A’=Jacobian of U’ and we
are dropping the subscripts k̄ on A, A’).
Now since (A0 )n ≈ (A0k )n ≈ (Pic V )n by 4.4.6 a the elements of (A0 )n can
be annihilated in some V 0 /V . Choose such a V’ and let A ⊂ (A0 )n be the image
of (A)n . We may assume A ∩ (A00 )n = 0. Now A is a G-submodule of (A0 )n .
Therefore, by 4.4.6 b applied to (A0 )n , G has trivial action on A, and H 1 (G, A) ≈
Hom(G, A) ≈ A (not canon.). Applying 4.4.6 b again, H 1 (G, A) ⊂ H 1 (G, (A0 )n ),
hence the image of H 1 (G, (A)n ) is exactly Hom(G, A). We are going to show
that the image of Z in H 1 (G, A) is zero at this stage.
Let k0 {x, y} be the henselization of oV,Q at some closed point Q of V (say
Q is a node of X) and let S be the ring (k0 {x, y})t (t = xy if x and y are chosen
85
suitably). We have a diagram
Uo spec S H 2 (U, µn ) / H 2 (spec S, µ )

n
O O
and hence
² ²
V o spec k0 {x, y} H 2 (V, µn ) / H 2 (spec k {x, y}, µ )
0 n 0
Therefore Z is mapped to zero in H 2 (spec S, µn ) for each node Q of X. Equiva-

lently, the image of Z in H 2 (S, µn ) must be zero for Q any node of X’ (the closed
fibre of V’).
The reader will verify that there is a morphism of spectral sequence
H p (G, H q (Ū 0 , µn )) +3 H ∗ (U 0 , µn )
² ²
H p (G, H q (spec S̄, µn )) +3 H ∗ (spec S, µn )
(Ū 0 = U ⊗, k̄, S̄ = S ⊗k k̄).

Applying 4.3.6 and the above discussion, one sees that in order to show image
of Z in H 1 (G, (A0 )n ) is zero, it suffices to show that if a ∈ A ⊂ H(Ū 0 , µn ) and
a 6= 0 then the image of a in H 1 (S̄, µn ) is not zero for some node Q of X’.
Let D be a divisor on V’ representing a in (Pic U )n and say D is chosen so
as to avoid the nodes of X. Then there is a rational function f on V’ so that (f ) =
P P
nD+ si Xi0 for some si ∈ Z, Xi0 the components of X’. Here si Xi0 is determined
P
mod X 0 and mod n, but since a 6∈ (A0 )n , si Xi0 6≡ 0( mod X 0 , n). Hence
for some pair of indices, say for 1, 2, s1 6≡ s2 ( mod n) and we may assume
X10 ∩ X20 6= ∅. Let Q be a point of intersection of X10 , X20 . f represents a unit in
the ring S obtained from Q, and in the notation of Section 4.3 (cf. 4.3.6) this
unit is not zero in (Z/n ⊕ Z/n)/∆ since s1 6≡ s2 ( mod n). Obviously this implies
that a does not have zero image in H 1 (S̄, µn ) (cf. 4.2.5 if necessary), and we are
done.
Q. E. D.
86
4.6 Case of an algebraic surface

Let π : V → C be a map of a complete nonsingular algebraic surface V onto
a nonsingular curve C, everything defined over an algebraically closed field k0 .
We assume that V /C has a section, that the generic fibre of π is geometrically
irreducible and simple, and that all fibres are reduced, with at most nodes as
singularities. We want to get information about the cohomology of V in terms of
π.
Given an algebraic surface V /k0 , it will in general be necessary to blow up
a few points of V in order that such a map exist, so one should first examine the
effect of a locally quadratic transformation f : V 0 → V at a closed point Q of V
on the cohomology. This is easily done, and one finds


 Gm q = 0

Rq f∗ Gm = ZQ q=1 (ignore p)


 0 q>1
 (4.6.1)

 µn
 q=0
Rq f∗ µn = (Z/n)Q q = 2 ((n, p) = 1)


 0 q=6 0, 2
where ZQ denotes the extension by zero of the sheaf Z on Q, and (Z/n)Q is

similar. Hence the only change in H q (V, Gm ) is in the picard group, where it is
the obvious one.
Returning to the map π, we have R0 π∗ Gm = GmC since V /C is proper, and
R1 π∗ Gm = Pic V /C is the relative picard functor (by definition, cf. [3] n◦ 232)
viewed as a sheaf on C. Since V /C has a section, H 0 R1 = Pic V /Pic C (cf. ibid).
Passing to the limit to compute the stalks of Rq π∗ Gm , and applying 4.2.2, 4.5.1
ii one finds


 Gm
 q=0
q
R π∗ Gm = Pic V /C q = 1 (ignore p) (4.6.2)


 0 q>1
Let i : c → C be the generic point, A/cthe Jacobian of the generic fibre U

of π. If F ⊂ Pic V /C denotes the subsheaf of divisor classes whose image on U
87
4.6 CASE OF AN ALGEBRAIC SURFACE
is of degree zero, we get exact sequence
0 → F → Pic V /C → Z → 0
(4.6.3)
0 → ² → F → i∗ A → 0
where ² is concentrated at the points of C whose fibre for π is reducible, and A is

viewed as a sheaf on c. So H q (C, ²) = 0, q > 0, hence H q (C, F ) ≈ H q (C, i∗ A), q >
0. Now H 1 (C, Z) = 0 (cf. Section 4.1). Moreover, since V /C has a section, the
map H 0 (C, Pic V /C) → H 0 (C, Z) obtained from the first sequence is surjective.
Therefore H 1 (C, Pic V /C) ≈ H 1 (C, i∗ A), whence by 4.6.2
H 2 (V, Gm ) ≈ H 1 (C, i∗ A) (ignore p) (4.6.4)
Applying 4.1.1
0 → Pic V /n → H 2 (V, µn ) → H 1 (C, i∗ A)n → 0 ((n, p) = 1) (4.6.5)
Remark 4. The image of Pic V in H 2 (V, µn ) is analogous to the part ρ of the

classical b2 (second betti number) generated by the algebraic cycles, hence ρ0 =
b2 − ρ is interpreted by 4.6.5 in terms of ”locally trivial principal homogeneous
spaces” of A (cf. Section 4.2 and [12]). The sequence 4.6.5 may be viewed as a
clarification of the results of [8]. In case V /C is a pencil of elliptic curves it is also
closely related to certain aspects of the beautiful paper of [9]. In order to show
that H 2 (V, µn ) has the expected value, it is at present necessary to use explicit
calculations of a somewhat devious nature involving the group H 1 (C, i∗ A). These
calculations may be found in [8] and [7].
To get information about H q (V, µn ) for q > 2, it is more convenient to

analyze the Leray spectral sequence
H p (V, Rq π∗ µn ) =⇒ H ∗ (V, µn )
directly. For simplicity we make a further assumption on π, namely that

all the fibres are irreducible. Then the sheaf ² of 4.6.3 is zero, and one sees
88
immediately that R1 π∗ µn ≈ (i∗ A)n ≈ i∗ (A)n . Moreover, there is an injection

(Pic V /C)/n → R2 π∗ µn . By 4.6.3, (Pic V /C)/n ≈ Z/n. Examining the stalks of
R2 π∗ µn by means of 4.2.2 and 4.5.1 i, one finds the injection is an isomorphism.
Hence


 µn q=0


 i (A) q=1 ((n, p) = 1, and the
∗ n
R q π ∗ µn = (4.6.6)

 Z/n q=2 fibres of π irreducible)



0 q>2
Utilizing the duality for finite group schemes over c (cf. Section 4.2) and the
autoduality of A, one finds the nonzero terms E2p,q of 4.6.1 are
\
Z/n(Pic C)n µ
cn
E2p,q = E∞
p,q
= ck )n )
(Ak )n H 1 (C, i∗ (A)n ) (A (4.6.7)
µn (Pic C)n Z/n
Here the transgressions H 0 R1 → H 2 R0 and H 0 R2 → H 2 R1 are zero. This is

seen by comparing 4.6.7 with the known values of the ending H q (V, µn ) (q=1,
2) of 4.6.1. For the second, use 4.6.5 and the so called ”Kummer sequence”
n
0 → i∗ (A)n → i∗ A → i∗ A → 0 which is exact if π has irreducible fibres.
89
4.6 CASE OF AN ALGEBRAIC SURFACE
90
BIBLIOGRAPHY
BIBLIOGRAPHY
[1] Grothendieck, A., Éléments de Géométrie Algébrique, Pub. Math. del Inst.
des Hautes Études Scient, 1960.
[2] Grothendieck, A., Séminaire de Géométrie Algébrique, des l’Inst. des Hautes
Études Scient, 1960-61.
[3] Séminaire Bourbaki - Textes des Conferences.
[4] Grothendieck, A., Sur Quelque Points d’Algebre Homologique, Tohoku Math.
Jour., IX, 1957, 119-221.
[5] Azumaya, G., on Maximally Central Algebras, Nagoya Math. Jour., 2, 1951.
[6] Auslander, M. and Goldman, O., The Brauer Group of a Commutative Ring,
Trans. Amer. Math. Soc., 97, n◦ 3, 1960.
[7] Artin, M., Some Numerical Criteria for Contractability of Curves on an

Algebraic Surface, Amer. Jour. (forthcoming).
[8] Igusa, J., Betti and Picard Numbers of Abstract Algebraic Surfaces, Proc.
Acad. Sci., 46, n◦ 5, 1960.
[9] Kodaira, K., On Compact Analytic Surfaces, Analytic Functions, Princeton,

1960.
[10] Lang, S., On Quasi Algebraic Closure, Annals of Math., 55, n◦ 2, 1952.
[11] Nagata, M. On the Theory of Henselian Rings, Nagoya Math. Jour. 5, 1953;
7, 1954 and Mem. Coll. Sci. Univ. Kyoto, 32, 1959-60.
[12] Ogg, A. Cohomology of Abelian Varieties over Function Fields, Annals of

Math. (forthcoming).
91

Artin - Grothendieck Topologies

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Artin - Grothendieck Topologies

Diunggah oleh

Hak Cipta:

Format Tersedia

GROTHENDIECK TOPOLOGIES

3 The étale Grothendieck topology for schemes 37

4 Calculations for curves and surfaces 69

4.4 Digression on the Picard group of a curve over k0 {t} . . . . . . . 76

Definition 1.1.1. A Grothendieck topology T consists of a category T and a set

1. If φi is an isomorphism then {φ} ∈ Cov T ;

3. If {Ui → U } ∈ Cov T and V → U ∈ T is arbitrary then Ui ×U V exists and

We will abuse language and call T a topology.

Definition 1.1.2. Let T be a topology and C a category with products. A

(S) If {Ui → U } ∈ Cov T then the diagram

We shall restrict our attention to presheaves and sheaves with value in Ab

TAUTOLOGY. Every representable functor on C is a sheaf of sets on TC .

Therefore the category of abelian sheaves on TG is equivalent with the cate-

Example 1.1.5. Let G be a profinite group and set TC = category of finite

Example 1.1.6. Let X be a nötherian scheme, and define TX by

TX = category of schemes Y/X étale, finite type

Cov TX = finite surjective families of maps

For instance, if X = spec k, k a field, then TX is dual to the category of

1.2 Generalities on lim

Proposition 1.2.1. If C = Sets (resp. Ab) then lim F is representable.

where R is the equivalence relation (resp. subgroup) generated by pairs (x, y)

j1 _> · · · jnbD (1.2.1)

(this doesn’t depend on C). In particular, if I is a discrete category (a category

The following axioms for I are useful:

L2 Given a diagram i ⇒ j, ∃ a map j → k such that the two maps i → k

Proposition 1.2.2. The functor F → Ab defined by F Ã lim F is right exact.

Proposition 1.2.3. Suppose I 6= ∅ and that (L1,2,3) hold in I. Then set

Proof: Suppose first (L1,3) hold, and let

is surjective, and hence set

Let x̄, ȳ ∈ set

Then there exist diagrams

in I with x = x0 in Fj ; y = y 0 in Fj 0 . Apply (L1) to the diagram

Corollary 1.2.4. If (L1,2) hold in I then lim is an exact functor: F → Ab.

One verified easily the following:

Hence ξ includes a map Fj → X and we are forced to define Fi → X by com-

Proposition 1.2.7. Let F, G, H : I → Sets be three functors and let F → G,

MORAL. Axioms (L1,2,3) make I as good as an inductive system for limits.

Proposition 1.2.8. Let I be given, F the category of functors I → Ab. For

Theorem 1.3.1 (Kan). There is a functor fp : P → P 0 left adjoint to f P is right

We sketch the proof for the convenience of the reader:

Ob IYf = {pairs(X, φ)|X ∈ ob C, φ ∈ HomC 0 (Y, f (X))}

(b) If F is a presheaf on C, we get a functor FY : IY◦ → Ab by FY ((X, φ)) =

fP F is made into a presheaf on C 0 (i.e. a functor) by (a). Because of (b), F Ã fP F

Corollary 1.3.2. If fP is exact then f P carries injectives into injectives.

For, if G ∈ P 0 is injective, i.e., Hom(x, G) is an exact functor for x ∈ P 0 we

Corollary 1.3.3. If F ∈ PC is injective then F (X) is an injective abelian group

For, apply 1.3.2.

Notation 1.3.1. Let i : {X} → C be as above. We write ZX = iP Z, so ZX ∈ PC

Corollary 1.3.4. Every F ∈ PC can be embedded in an injective.

Ob IYf = {(X, φ)|X ∈ C, φ : f (X) ← Y }

By adjointness of g to f this is the same as {(X, ψ)|X ∈ C, ψ : X ← g(Y )}.

1.4 Čech cohomology

If F is a presheaf on C we obtain a diagram

in the canonical way. Define

and verify dn+1 ◦ dn = 0. One obtains a functor PC →(cochain complexes) which

Proof of Theorem 1.4.1: Suppose F ∈ PC is injective and let ZX ∈ PC be as in

defined above is exact. This is the same as showing

= Hom(⊕ZUα , F ) / Hom(⊕ZU × U , F ) / ···

⊕ZUα o ⊕ZUα ×V Uβ o ···

⊕ZUα (Y ) o ⊕ZUα ×V Uβ (Y ) o ··· (1.4.1)

Hence we need to show that a diagram of the form

n(i0 , · · · , in ) Ã n(l, i0 , · · · , in )(l ∈ S fixed)

for n(i0 , · · · , in ) = n in the (i0 , · · · , in )-th component of ⊕S n+1 Z. This completes